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
  42 #define LINKCHANGE_INT (2 * HZ)
  43 #define VF_TAKEOVER_INT (HZ / 10)
  44
  45 static unsigned int ring_size __ro_after_init = 128;
  46 module_param(ring_size, uint, 0444);
  47 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
  48 unsigned int netvsc_ring_bytes __ro_after_init;
  49
  50 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
  51                                 NETIF_MSG_LINK | NETIF_MSG_IFUP |
  52                                 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
  53                                 NETIF_MSG_TX_ERR;
  54
  55 static int debug = -1;
  56 module_param(debug, int, 0444);
  57 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  58
  59 static LIST_HEAD(netvsc_dev_list);
  60
  61 static void netvsc_change_rx_flags(struct net_device *net, int change)
  62 {
  63         struct net_device_context *ndev_ctx = netdev_priv(net);
  64         struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
  65         int inc;
  66
  67         if (!vf_netdev)
  68                 return;
  69
  70         if (change & IFF_PROMISC) {
  71                 inc = (net->flags & IFF_PROMISC) ? 1 : -1;
  72                 dev_set_promiscuity(vf_netdev, inc);
  73         }
  74
  75         if (change & IFF_ALLMULTI) {
  76                 inc = (net->flags & IFF_ALLMULTI) ? 1 : -1;
  77                 dev_set_allmulti(vf_netdev, inc);
  78         }
  79 }
  80
  81 static void netvsc_set_rx_mode(struct net_device *net)
  82 {
  83         struct net_device_context *ndev_ctx = netdev_priv(net);
  84         struct net_device *vf_netdev;
  85         struct netvsc_device *nvdev;
  86
  87         rcu_read_lock();
  88         vf_netdev = rcu_dereference(ndev_ctx->vf_netdev);
  89         if (vf_netdev) {
  90                 dev_uc_sync(vf_netdev, net);
  91                 dev_mc_sync(vf_netdev, net);
  92         }
  93
  94         nvdev = rcu_dereference(ndev_ctx->nvdev);
  95         if (nvdev)
  96                 rndis_filter_update(nvdev);
  97         rcu_read_unlock();
  98 }
  99
 100 static void netvsc_tx_enable(struct netvsc_device *nvscdev,
 101                              struct net_device *ndev)
 102 {
 103         nvscdev->tx_disable = false;
 104         virt_wmb(); /* ensure queue wake up mechanism is on */
 105
 106         netif_tx_wake_all_queues(ndev);
 107 }
 108
 109 static int netvsc_open(struct net_device *net)
 110 {
 111         struct net_device_context *ndev_ctx = netdev_priv(net);
 112         struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
 113         struct netvsc_device *nvdev = rtnl_dereference(ndev_ctx->nvdev);
 114         struct rndis_device *rdev;
 115         int ret = 0;
 116
 117         netif_carrier_off(net);
 118
 119         /* Open up the device */
 120         ret = rndis_filter_open(nvdev);
 121         if (ret != 0) {
 122                 netdev_err(net, "unable to open device (ret %d).\n", ret);
 123                 return ret;
 124         }
 125
 126         rdev = nvdev->extension;
 127         if (!rdev->link_state) {
 128                 netif_carrier_on(net);
 129                 netvsc_tx_enable(nvdev, net);
 130         }
 131
 132         if (vf_netdev) {
 133                 /* Setting synthetic device up transparently sets
 134                  * slave as up. If open fails, then slave will be
 135                  * still be offline (and not used).
 136                  */
 137                 ret = dev_open(vf_netdev, NULL);
 138                 if (ret)
 139                         netdev_warn(net,
 140                                     "unable to open slave: %s: %d\n",
 141                                     vf_netdev->name, ret);
 142         }
 143         return 0;
 144 }
 145
 146 static int netvsc_wait_until_empty(struct netvsc_device *nvdev)
 147 {
 148         unsigned int retry = 0;
 149         int i;
 150
 151         /* Ensure pending bytes in ring are read */
 152         for (;;) {
 153                 u32 aread = 0;
 154
 155                 for (i = 0; i < nvdev->num_chn; i++) {
 156                         struct vmbus_channel *chn
 157                                 = nvdev->chan_table[i].channel;
 158
 159                         if (!chn)
 160                                 continue;
 161
 162                         /* make sure receive not running now */
 163                         napi_synchronize(&nvdev->chan_table[i].napi);
 164
 165                         aread = hv_get_bytes_to_read(&chn->inbound);
 166                         if (aread)
 167                                 break;
 168
 169                         aread = hv_get_bytes_to_read(&chn->outbound);
 170                         if (aread)
 171                                 break;
 172                 }
 173
 174                 if (aread == 0)
 175                         return 0;
 176
 177                 if (++retry > RETRY_MAX)
 178                         return -ETIMEDOUT;
 179
 180                 usleep_range(RETRY_US_LO, RETRY_US_HI);
 181         }
 182 }
 183
 184 static void netvsc_tx_disable(struct netvsc_device *nvscdev,
 185                               struct net_device *ndev)
 186 {
 187         if (nvscdev) {
 188                 nvscdev->tx_disable = true;
 189                 virt_wmb(); /* ensure txq will not wake up after stop */
 190         }
 191
 192         netif_tx_disable(ndev);
 193 }
 194
 195 static int netvsc_close(struct net_device *net)
 196 {
 197         struct net_device_context *net_device_ctx = netdev_priv(net);
 198         struct net_device *vf_netdev
 199                 = rtnl_dereference(net_device_ctx->vf_netdev);
 200         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
 201         int ret;
 202
 203         netvsc_tx_disable(nvdev, net);
 204
 205         /* No need to close rndis filter if it is removed already */
 206         if (!nvdev)
 207                 return 0;
 208
 209         ret = rndis_filter_close(nvdev);
 210         if (ret != 0) {
 211                 netdev_err(net, "unable to close device (ret %d).\n", ret);
 212                 return ret;
 213         }
 214
 215         ret = netvsc_wait_until_empty(nvdev);
 216         if (ret)
 217                 netdev_err(net, "Ring buffer not empty after closing rndis\n");
 218
 219         if (vf_netdev)
 220                 dev_close(vf_netdev);
 221
 222         return ret;
 223 }
 224
 225 static inline void *init_ppi_data(struct rndis_message *msg,
 226                                   u32 ppi_size, u32 pkt_type)
 227 {
 228         struct rndis_packet *rndis_pkt = &msg->msg.pkt;
 229         struct rndis_per_packet_info *ppi;
 230
 231         rndis_pkt->data_offset += ppi_size;
 232         ppi = (void *)rndis_pkt + rndis_pkt->per_pkt_info_offset
 233                 + rndis_pkt->per_pkt_info_len;
 234
 235         ppi->size = ppi_size;
 236         ppi->type = pkt_type;
 237         ppi->internal = 0;
 238         ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
 239
 240         rndis_pkt->per_pkt_info_len += ppi_size;
 241
 242         return ppi + 1;
 243 }
 244
 245 /* Azure hosts don't support non-TCP port numbers in hashing for fragmented
 246  * packets. We can use ethtool to change UDP hash level when necessary.
 247  */
 248 static inline u32 netvsc_get_hash(
 249         struct sk_buff *skb,
 250         const struct net_device_context *ndc)
 251 {
 252         struct flow_keys flow;
 253         u32 hash, pkt_proto = 0;
 254         static u32 hashrnd __read_mostly;
 255
 256         net_get_random_once(&hashrnd, sizeof(hashrnd));
 257
 258         if (!skb_flow_dissect_flow_keys(skb, &flow, 0))
 259                 return 0;
 260
 261         switch (flow.basic.ip_proto) {
 262         case IPPROTO_TCP:
 263                 if (flow.basic.n_proto == htons(ETH_P_IP))
 264                         pkt_proto = HV_TCP4_L4HASH;
 265                 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
 266                         pkt_proto = HV_TCP6_L4HASH;
 267
 268                 break;
 269
 270         case IPPROTO_UDP:
 271                 if (flow.basic.n_proto == htons(ETH_P_IP))
 272                         pkt_proto = HV_UDP4_L4HASH;
 273                 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
 274                         pkt_proto = HV_UDP6_L4HASH;
 275
 276                 break;
 277         }
 278
 279         if (pkt_proto & ndc->l4_hash) {
 280                 return skb_get_hash(skb);
 281         } else {
 282                 if (flow.basic.n_proto == htons(ETH_P_IP))
 283                         hash = jhash2((u32 *)&flow.addrs.v4addrs, 2, hashrnd);
 284                 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
 285                         hash = jhash2((u32 *)&flow.addrs.v6addrs, 8, hashrnd);
 286                 else
 287                         return 0;
 288
 289                 __skb_set_sw_hash(skb, hash, false);
 290         }
 291
 292         return hash;
 293 }
 294
 295 static inline int netvsc_get_tx_queue(struct net_device *ndev,
 296                                       struct sk_buff *skb, int old_idx)
 297 {
 298         const struct net_device_context *ndc = netdev_priv(ndev);
 299         struct sock *sk = skb->sk;
 300         int q_idx;
 301
 302         q_idx = ndc->tx_table[netvsc_get_hash(skb, ndc) &
 303                               (VRSS_SEND_TAB_SIZE - 1)];
 304
 305         /* If queue index changed record the new value */
 306         if (q_idx != old_idx &&
 307             sk && sk_fullsock(sk) && rcu_access_pointer(sk->sk_dst_cache))
 308                 sk_tx_queue_set(sk, q_idx);
 309
 310         return q_idx;
 311 }
 312
 313 /*
 314  * Select queue for transmit.
 315  *
 316  * If a valid queue has already been assigned, then use that.
 317  * Otherwise compute tx queue based on hash and the send table.
 318  *
 319  * This is basically similar to default (netdev_pick_tx) with the added step
 320  * of using the host send_table when no other queue has been assigned.
 321  *
 322  * TODO support XPS - but get_xps_queue not exported
 323  */
 324 static u16 netvsc_pick_tx(struct net_device *ndev, struct sk_buff *skb)
 325 {
 326         int q_idx = sk_tx_queue_get(skb->sk);
 327
 328         if (q_idx < 0 || skb->ooo_okay || q_idx >= ndev->real_num_tx_queues) {
 329                 /* If forwarding a packet, we use the recorded queue when
 330                  * available for better cache locality.
 331                  */
 332                 if (skb_rx_queue_recorded(skb))
 333                         q_idx = skb_get_rx_queue(skb);
 334                 else
 335                         q_idx = netvsc_get_tx_queue(ndev, skb, q_idx);
 336         }
 337
 338         return q_idx;
 339 }
 340
 341 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
 342                                struct net_device *sb_dev)
 343 {
 344         struct net_device_context *ndc = netdev_priv(ndev);
 345         struct net_device *vf_netdev;
 346         u16 txq;
 347
 348         rcu_read_lock();
 349         vf_netdev = rcu_dereference(ndc->vf_netdev);
 350         if (vf_netdev) {
 351                 const struct net_device_ops *vf_ops = vf_netdev->netdev_ops;
 352
 353                 if (vf_ops->ndo_select_queue)
 354                         txq = vf_ops->ndo_select_queue(vf_netdev, skb, sb_dev);
 355                 else
 356                         txq = netdev_pick_tx(vf_netdev, skb, NULL);
 357
 358                 /* Record the queue selected by VF so that it can be
 359                  * used for common case where VF has more queues than
 360                  * the synthetic device.
 361                  */
 362                 qdisc_skb_cb(skb)->slave_dev_queue_mapping = txq;
 363         } else {
 364                 txq = netvsc_pick_tx(ndev, skb);
 365         }
 366         rcu_read_unlock();
 367
 368         while (txq >= ndev->real_num_tx_queues)
 369                 txq -= ndev->real_num_tx_queues;
 370
 371         return txq;
 372 }
 373
 374 static u32 fill_pg_buf(unsigned long hvpfn, u32 offset, u32 len,
 375                        struct hv_page_buffer *pb)
 376 {
 377         int j = 0;
 378
 379         hvpfn += offset >> HV_HYP_PAGE_SHIFT;
 380         offset = offset & ~HV_HYP_PAGE_MASK;
 381
 382         while (len > 0) {
 383                 unsigned long bytes;
 384
 385                 bytes = HV_HYP_PAGE_SIZE - offset;
 386                 if (bytes > len)
 387                         bytes = len;
 388                 pb[j].pfn = hvpfn;
 389                 pb[j].offset = offset;
 390                 pb[j].len = bytes;
 391
 392                 offset += bytes;
 393                 len -= bytes;
 394
 395                 if (offset == HV_HYP_PAGE_SIZE && len) {
 396                         hvpfn++;
 397                         offset = 0;
 398                         j++;
 399                 }
 400         }
 401
 402         return j + 1;
 403 }
 404
 405 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
 406                            struct hv_netvsc_packet *packet,
 407                            struct hv_page_buffer *pb)
 408 {
 409         u32 slots_used = 0;
 410         char *data = skb->data;
 411         int frags = skb_shinfo(skb)->nr_frags;
 412         int i;
 413
 414         /* The packet is laid out thus:
 415          * 1. hdr: RNDIS header and PPI
 416          * 2. skb linear data
 417          * 3. skb fragment data
 418          */
 419         slots_used += fill_pg_buf(virt_to_hvpfn(hdr),
 420                                   offset_in_hvpage(hdr),
 421                                   len,
 422                                   &pb[slots_used]);
 423
 424         packet->rmsg_size = len;
 425         packet->rmsg_pgcnt = slots_used;
 426
 427         slots_used += fill_pg_buf(virt_to_hvpfn(data),
 428                                   offset_in_hvpage(data),
 429                                   skb_headlen(skb),
 430                                   &pb[slots_used]);
 431
 432         for (i = 0; i < frags; i++) {
 433                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 434
 435                 slots_used += fill_pg_buf(page_to_hvpfn(skb_frag_page(frag)),
 436                                           skb_frag_off(frag),
 437                                           skb_frag_size(frag),
 438                                           &pb[slots_used]);
 439         }
 440         return slots_used;
 441 }
 442
 443 static int count_skb_frag_slots(struct sk_buff *skb)
 444 {
 445         int i, frags = skb_shinfo(skb)->nr_frags;
 446         int pages = 0;
 447
 448         for (i = 0; i < frags; i++) {
 449                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 450                 unsigned long size = skb_frag_size(frag);
 451                 unsigned long offset = skb_frag_off(frag);
 452
 453                 /* Skip unused frames from start of page */
 454                 offset &= ~HV_HYP_PAGE_MASK;
 455                 pages += HVPFN_UP(offset + size);
 456         }
 457         return pages;
 458 }
 459
 460 static int netvsc_get_slots(struct sk_buff *skb)
 461 {
 462         char *data = skb->data;
 463         unsigned int offset = offset_in_hvpage(data);
 464         unsigned int len = skb_headlen(skb);
 465         int slots;
 466         int frag_slots;
 467
 468         slots = DIV_ROUND_UP(offset + len, HV_HYP_PAGE_SIZE);
 469         frag_slots = count_skb_frag_slots(skb);
 470         return slots + frag_slots;
 471 }
 472
 473 static u32 net_checksum_info(struct sk_buff *skb)
 474 {
 475         if (skb->protocol == htons(ETH_P_IP)) {
 476                 struct iphdr *ip = ip_hdr(skb);
 477
 478                 if (ip->protocol == IPPROTO_TCP)
 479                         return TRANSPORT_INFO_IPV4_TCP;
 480                 else if (ip->protocol == IPPROTO_UDP)
 481                         return TRANSPORT_INFO_IPV4_UDP;
 482         } else {
 483                 struct ipv6hdr *ip6 = ipv6_hdr(skb);
 484
 485                 if (ip6->nexthdr == IPPROTO_TCP)
 486                         return TRANSPORT_INFO_IPV6_TCP;
 487                 else if (ip6->nexthdr == IPPROTO_UDP)
 488                         return TRANSPORT_INFO_IPV6_UDP;
 489         }
 490
 491         return TRANSPORT_INFO_NOT_IP;
 492 }
 493
 494 /* Send skb on the slave VF device. */
 495 static int netvsc_vf_xmit(struct net_device *net, struct net_device *vf_netdev,
 496                           struct sk_buff *skb)
 497 {
 498         struct net_device_context *ndev_ctx = netdev_priv(net);
 499         unsigned int len = skb->len;
 500         int rc;
 501
 502         skb->dev = vf_netdev;
 503         skb_record_rx_queue(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
 504
 505         rc = dev_queue_xmit(skb);
 506         if (likely(rc == NET_XMIT_SUCCESS || rc == NET_XMIT_CN)) {
 507                 struct netvsc_vf_pcpu_stats *pcpu_stats
 508                         = this_cpu_ptr(ndev_ctx->vf_stats);
 509
 510                 u64_stats_update_begin(&pcpu_stats->syncp);
 511                 pcpu_stats->tx_packets++;
 512                 pcpu_stats->tx_bytes += len;
 513                 u64_stats_update_end(&pcpu_stats->syncp);
 514         } else {
 515                 this_cpu_inc(ndev_ctx->vf_stats->tx_dropped);
 516         }
 517
 518         return rc;
 519 }
 520
 521 static int netvsc_xmit(struct sk_buff *skb, struct net_device *net, bool xdp_tx)
 522 {
 523         struct net_device_context *net_device_ctx = netdev_priv(net);
 524         struct hv_netvsc_packet *packet = NULL;
 525         int ret;
 526         unsigned int num_data_pgs;
 527         struct rndis_message *rndis_msg;
 528         struct net_device *vf_netdev;
 529         u32 rndis_msg_size;
 530         u32 hash;
 531         struct hv_page_buffer pb[MAX_PAGE_BUFFER_COUNT];
 532
 533         /* If VF is present and up then redirect packets to it.
 534          * Skip the VF if it is marked down or has no carrier.
 535          * If netpoll is in uses, then VF can not be used either.
 536          */
 537         vf_netdev = rcu_dereference_bh(net_device_ctx->vf_netdev);
 538         if (vf_netdev && netif_running(vf_netdev) &&
 539             netif_carrier_ok(vf_netdev) && !netpoll_tx_running(net) &&
 540             net_device_ctx->data_path_is_vf)
 541                 return netvsc_vf_xmit(net, vf_netdev, skb);
 542
 543         /* We will atmost need two pages to describe the rndis
 544          * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
 545          * of pages in a single packet. If skb is scattered around
 546          * more pages we try linearizing it.
 547          */
 548
 549         num_data_pgs = netvsc_get_slots(skb) + 2;
 550
 551         if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
 552                 ++net_device_ctx->eth_stats.tx_scattered;
 553
 554                 if (skb_linearize(skb))
 555                         goto no_memory;
 556
 557                 num_data_pgs = netvsc_get_slots(skb) + 2;
 558                 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
 559                         ++net_device_ctx->eth_stats.tx_too_big;
 560                         goto drop;
 561                 }
 562         }
 563
 564         /*
 565          * Place the rndis header in the skb head room and
 566          * the skb->cb will be used for hv_netvsc_packet
 567          * structure.
 568          */
 569         ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
 570         if (ret)
 571                 goto no_memory;
 572
 573         /* Use the skb control buffer for building up the packet */
 574         BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
 575                         sizeof_field(struct sk_buff, cb));
 576         packet = (struct hv_netvsc_packet *)skb->cb;
 577
 578         packet->q_idx = skb_get_queue_mapping(skb);
 579
 580         packet->total_data_buflen = skb->len;
 581         packet->total_bytes = skb->len;
 582         packet->total_packets = 1;
 583
 584         rndis_msg = (struct rndis_message *)skb->head;
 585
 586         /* Add the rndis header */
 587         rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
 588         rndis_msg->msg_len = packet->total_data_buflen;
 589
 590         rndis_msg->msg.pkt = (struct rndis_packet) {
 591                 .data_offset = sizeof(struct rndis_packet),
 592                 .data_len = packet->total_data_buflen,
 593                 .per_pkt_info_offset = sizeof(struct rndis_packet),
 594         };
 595
 596         rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
 597
 598         hash = skb_get_hash_raw(skb);
 599         if (hash != 0 && net->real_num_tx_queues > 1) {
 600                 u32 *hash_info;
 601
 602                 rndis_msg_size += NDIS_HASH_PPI_SIZE;
 603                 hash_info = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
 604                                           NBL_HASH_VALUE);
 605                 *hash_info = hash;
 606         }
 607
 608         /* When using AF_PACKET we need to drop VLAN header from
 609          * the frame and update the SKB to allow the HOST OS
 610          * to transmit the 802.1Q packet
 611          */
 612         if (skb->protocol == htons(ETH_P_8021Q)) {
 613                 u16 vlan_tci;
 614
 615                 skb_reset_mac_header(skb);
 616                 if (eth_type_vlan(eth_hdr(skb)->h_proto)) {
 617                         if (unlikely(__skb_vlan_pop(skb, &vlan_tci) != 0)) {
 618                                 ++net_device_ctx->eth_stats.vlan_error;
 619                                 goto drop;
 620                         }
 621
 622                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci);
 623                         /* Update the NDIS header pkt lengths */
 624                         packet->total_data_buflen -= VLAN_HLEN;
 625                         packet->total_bytes -= VLAN_HLEN;
 626                         rndis_msg->msg_len = packet->total_data_buflen;
 627                         rndis_msg->msg.pkt.data_len = packet->total_data_buflen;
 628                 }
 629         }
 630
 631         if (skb_vlan_tag_present(skb)) {
 632                 struct ndis_pkt_8021q_info *vlan;
 633
 634                 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
 635                 vlan = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
 636                                      IEEE_8021Q_INFO);
 637
 638                 vlan->value = 0;
 639                 vlan->vlanid = skb_vlan_tag_get_id(skb);
 640                 vlan->cfi = skb_vlan_tag_get_cfi(skb);
 641                 vlan->pri = skb_vlan_tag_get_prio(skb);
 642         }
 643
 644         if (skb_is_gso(skb)) {
 645                 struct ndis_tcp_lso_info *lso_info;
 646
 647                 rndis_msg_size += NDIS_LSO_PPI_SIZE;
 648                 lso_info = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
 649                                          TCP_LARGESEND_PKTINFO);
 650
 651                 lso_info->value = 0;
 652                 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
 653                 if (skb->protocol == htons(ETH_P_IP)) {
 654                         lso_info->lso_v2_transmit.ip_version =
 655                                 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
 656                         ip_hdr(skb)->tot_len = 0;
 657                         ip_hdr(skb)->check = 0;
 658                         tcp_hdr(skb)->check =
 659                                 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
 660                                                    ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 661                 } else {
 662                         lso_info->lso_v2_transmit.ip_version =
 663                                 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
 664                         tcp_v6_gso_csum_prep(skb);
 665                 }
 666                 lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb);
 667                 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
 668         } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
 669                 if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) {
 670                         struct ndis_tcp_ip_checksum_info *csum_info;
 671
 672                         rndis_msg_size += NDIS_CSUM_PPI_SIZE;
 673                         csum_info = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
 674                                                   TCPIP_CHKSUM_PKTINFO);
 675
 676                         csum_info->value = 0;
 677                         csum_info->transmit.tcp_header_offset = skb_transport_offset(skb);
 678
 679                         if (skb->protocol == htons(ETH_P_IP)) {
 680                                 csum_info->transmit.is_ipv4 = 1;
 681
 682                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
 683                                         csum_info->transmit.tcp_checksum = 1;
 684                                 else
 685                                         csum_info->transmit.udp_checksum = 1;
 686                         } else {
 687                                 csum_info->transmit.is_ipv6 = 1;
 688
 689                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
 690                                         csum_info->transmit.tcp_checksum = 1;
 691                                 else
 692                                         csum_info->transmit.udp_checksum = 1;
 693                         }
 694                 } else {
 695                         /* Can't do offload of this type of checksum */
 696                         if (skb_checksum_help(skb))
 697                                 goto drop;
 698                 }
 699         }
 700
 701         /* Start filling in the page buffers with the rndis hdr */
 702         rndis_msg->msg_len += rndis_msg_size;
 703         packet->total_data_buflen = rndis_msg->msg_len;
 704         packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
 705                                                skb, packet, pb);
 706
 707         /* timestamp packet in software */
 708         skb_tx_timestamp(skb);
 709
 710         ret = netvsc_send(net, packet, rndis_msg, pb, skb, xdp_tx);
 711         if (likely(ret == 0))
 712                 return NETDEV_TX_OK;
 713
 714         if (ret == -EAGAIN) {
 715                 ++net_device_ctx->eth_stats.tx_busy;
 716                 return NETDEV_TX_BUSY;
 717         }
 718
 719         if (ret == -ENOSPC)
 720                 ++net_device_ctx->eth_stats.tx_no_space;
 721
 722 drop:
 723         dev_kfree_skb_any(skb);
 724         net->stats.tx_dropped++;
 725
 726         return NETDEV_TX_OK;
 727
 728 no_memory:
 729         ++net_device_ctx->eth_stats.tx_no_memory;
 730         goto drop;
 731 }
 732
 733 static netdev_tx_t netvsc_start_xmit(struct sk_buff *skb,
 734                                      struct net_device *ndev)
 735 {
 736         return netvsc_xmit(skb, ndev, false);
 737 }
 738
 739 /*
 740  * netvsc_linkstatus_callback - Link up/down notification
 741  */
 742 void netvsc_linkstatus_callback(struct net_device *net,
 743                                 struct rndis_message *resp,
 744                                 void *data, u32 data_buflen)
 745 {
 746         struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
 747         struct net_device_context *ndev_ctx = netdev_priv(net);
 748         struct netvsc_reconfig *event;
 749         unsigned long flags;
 750
 751         /* Ensure the packet is big enough to access its fields */
 752         if (resp->msg_len - RNDIS_HEADER_SIZE < sizeof(struct rndis_indicate_status)) {
 753                 netdev_err(net, "invalid rndis_indicate_status packet, len: %u\n",
 754                            resp->msg_len);
 755                 return;
 756         }
 757
 758         /* Copy the RNDIS indicate status into nvchan->recv_buf */
 759         memcpy(indicate, data + RNDIS_HEADER_SIZE, sizeof(*indicate));
 760
 761         /* Update the physical link speed when changing to another vSwitch */
 762         if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
 763                 u32 speed;
 764
 765                 /* Validate status_buf_offset and status_buflen.
 766                  *
 767                  * Certain (pre-Fe) implementations of Hyper-V's vSwitch didn't account
 768                  * for the status buffer field in resp->msg_len; perform the validation
 769                  * using data_buflen (>= resp->msg_len).
 770                  */
 771                 if (indicate->status_buflen < sizeof(speed) ||
 772                     indicate->status_buf_offset < sizeof(*indicate) ||
 773                     data_buflen - RNDIS_HEADER_SIZE < indicate->status_buf_offset ||
 774                     data_buflen - RNDIS_HEADER_SIZE - indicate->status_buf_offset
 775                                 < indicate->status_buflen) {
 776                         netdev_err(net, "invalid rndis_indicate_status packet\n");
 777                         return;
 778                 }
 779
 780                 speed = *(u32 *)(data + RNDIS_HEADER_SIZE + indicate->status_buf_offset) / 10000;
 781                 ndev_ctx->speed = speed;
 782                 return;
 783         }
 784
 785         /* Handle these link change statuses below */
 786         if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
 787             indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
 788             indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
 789                 return;
 790
 791         if (net->reg_state != NETREG_REGISTERED)
 792                 return;
 793
 794         event = kzalloc(sizeof(*event), GFP_ATOMIC);
 795         if (!event)
 796                 return;
 797         event->event = indicate->status;
 798
 799         spin_lock_irqsave(&ndev_ctx->lock, flags);
 800         list_add_tail(&event->list, &ndev_ctx->reconfig_events);
 801         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
 802
 803         schedule_delayed_work(&ndev_ctx->dwork, 0);
 804 }
 805
 806 static void netvsc_xdp_xmit(struct sk_buff *skb, struct net_device *ndev)
 807 {
 808         int rc;
 809
 810         skb->queue_mapping = skb_get_rx_queue(skb);
 811         __skb_push(skb, ETH_HLEN);
 812
 813         rc = netvsc_xmit(skb, ndev, true);
 814
 815         if (dev_xmit_complete(rc))
 816                 return;
 817
 818         dev_kfree_skb_any(skb);
 819         ndev->stats.tx_dropped++;
 820 }
 821
 822 static void netvsc_comp_ipcsum(struct sk_buff *skb)
 823 {
 824         struct iphdr *iph = (struct iphdr *)skb->data;
 825
 826         iph->check = 0;
 827         iph->check = ip_fast_csum(iph, iph->ihl);
 828 }
 829
 830 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
 831                                              struct netvsc_channel *nvchan,
 832                                              struct xdp_buff *xdp)
 833 {
 834         struct napi_struct *napi = &nvchan->napi;
 835         const struct ndis_pkt_8021q_info *vlan = &nvchan->rsc.vlan;
 836         const struct ndis_tcp_ip_checksum_info *csum_info =
 837                                                 &nvchan->rsc.csum_info;
 838         const u32 *hash_info = &nvchan->rsc.hash_info;
 839         u8 ppi_flags = nvchan->rsc.ppi_flags;
 840         struct sk_buff *skb;
 841         void *xbuf = xdp->data_hard_start;
 842         int i;
 843
 844         if (xbuf) {
 845                 unsigned int hdroom = xdp->data - xdp->data_hard_start;
 846                 unsigned int xlen = xdp->data_end - xdp->data;
 847                 unsigned int frag_size = xdp->frame_sz;
 848
 849                 skb = build_skb(xbuf, frag_size);
 850
 851                 if (!skb) {
 852                         __free_page(virt_to_page(xbuf));
 853                         return NULL;
 854                 }
 855
 856                 skb_reserve(skb, hdroom);
 857                 skb_put(skb, xlen);
 858                 skb->dev = napi->dev;
 859         } else {
 860                 skb = napi_alloc_skb(napi, nvchan->rsc.pktlen);
 861
 862                 if (!skb)
 863                         return NULL;
 864
 865                 /* Copy to skb. This copy is needed here since the memory
 866                  * pointed by hv_netvsc_packet cannot be deallocated.
 867                  */
 868                 for (i = 0; i < nvchan->rsc.cnt; i++)
 869                         skb_put_data(skb, nvchan->rsc.data[i],
 870                                      nvchan->rsc.len[i]);
 871         }
 872
 873         skb->protocol = eth_type_trans(skb, net);
 874
 875         /* skb is already created with CHECKSUM_NONE */
 876         skb_checksum_none_assert(skb);
 877
 878         /* Incoming packets may have IP header checksum verified by the host.
 879          * They may not have IP header checksum computed after coalescing.
 880          * We compute it here if the flags are set, because on Linux, the IP
 881          * checksum is always checked.
 882          */
 883         if ((ppi_flags & NVSC_RSC_CSUM_INFO) && csum_info->receive.ip_checksum_value_invalid &&
 884             csum_info->receive.ip_checksum_succeeded &&
 885             skb->protocol == htons(ETH_P_IP)) {
 886                 /* Check that there is enough space to hold the IP header. */
 887                 if (skb_headlen(skb) < sizeof(struct iphdr)) {
 888                         kfree_skb(skb);
 889                         return NULL;
 890                 }
 891                 netvsc_comp_ipcsum(skb);
 892         }
 893
 894         /* Do L4 checksum offload if enabled and present. */
 895         if ((ppi_flags & NVSC_RSC_CSUM_INFO) && (net->features & NETIF_F_RXCSUM)) {
 896                 if (csum_info->receive.tcp_checksum_succeeded ||
 897                     csum_info->receive.udp_checksum_succeeded)
 898                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 899         }
 900
 901         if ((ppi_flags & NVSC_RSC_HASH_INFO) && (net->features & NETIF_F_RXHASH))
 902                 skb_set_hash(skb, *hash_info, PKT_HASH_TYPE_L4);
 903
 904         if (ppi_flags & NVSC_RSC_VLAN) {
 905                 u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT) |
 906                         (vlan->cfi ? VLAN_CFI_MASK : 0);
 907
 908                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 909                                        vlan_tci);
 910         }
 911
 912         return skb;
 913 }
 914
 915 /*
 916  * netvsc_recv_callback -  Callback when we receive a packet from the
 917  * "wire" on the specified device.
 918  */
 919 int netvsc_recv_callback(struct net_device *net,
 920                          struct netvsc_device *net_device,
 921                          struct netvsc_channel *nvchan)
 922 {
 923         struct net_device_context *net_device_ctx = netdev_priv(net);
 924         struct vmbus_channel *channel = nvchan->channel;
 925         u16 q_idx = channel->offermsg.offer.sub_channel_index;
 926         struct sk_buff *skb;
 927         struct netvsc_stats *rx_stats = &nvchan->rx_stats;
 928         struct xdp_buff xdp;
 929         u32 act;
 930
 931         if (net->reg_state != NETREG_REGISTERED)
 932                 return NVSP_STAT_FAIL;
 933
 934         act = netvsc_run_xdp(net, nvchan, &xdp);
 935
 936         if (act != XDP_PASS && act != XDP_TX) {
 937                 u64_stats_update_begin(&rx_stats->syncp);
 938                 rx_stats->xdp_drop++;
 939                 u64_stats_update_end(&rx_stats->syncp);
 940
 941                 return NVSP_STAT_SUCCESS; /* consumed by XDP */
 942         }
 943
 944         /* Allocate a skb - TODO direct I/O to pages? */
 945         skb = netvsc_alloc_recv_skb(net, nvchan, &xdp);
 946
 947         if (unlikely(!skb)) {
 948                 ++net_device_ctx->eth_stats.rx_no_memory;
 949                 return NVSP_STAT_FAIL;
 950         }
 951
 952         skb_record_rx_queue(skb, q_idx);
 953
 954         /*
 955          * Even if injecting the packet, record the statistics
 956          * on the synthetic device because modifying the VF device
 957          * statistics will not work correctly.
 958          */
 959         u64_stats_update_begin(&rx_stats->syncp);
 960         rx_stats->packets++;
 961         rx_stats->bytes += nvchan->rsc.pktlen;
 962
 963         if (skb->pkt_type == PACKET_BROADCAST)
 964                 ++rx_stats->broadcast;
 965         else if (skb->pkt_type == PACKET_MULTICAST)
 966                 ++rx_stats->multicast;
 967         u64_stats_update_end(&rx_stats->syncp);
 968
 969         if (act == XDP_TX) {
 970                 netvsc_xdp_xmit(skb, net);
 971                 return NVSP_STAT_SUCCESS;
 972         }
 973
 974         napi_gro_receive(&nvchan->napi, skb);
 975         return NVSP_STAT_SUCCESS;
 976 }
 977
 978 static void netvsc_get_drvinfo(struct net_device *net,
 979                                struct ethtool_drvinfo *info)
 980 {
 981         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
 982         strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
 983 }
 984
 985 static void netvsc_get_channels(struct net_device *net,
 986                                 struct ethtool_channels *channel)
 987 {
 988         struct net_device_context *net_device_ctx = netdev_priv(net);
 989         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
 990
 991         if (nvdev) {
 992                 channel->max_combined   = nvdev->max_chn;
 993                 channel->combined_count = nvdev->num_chn;
 994         }
 995 }
 996
 997 /* Alloc struct netvsc_device_info, and initialize it from either existing
 998  * struct netvsc_device, or from default values.
 999  */
1000 static
1001 struct netvsc_device_info *netvsc_devinfo_get(struct netvsc_device *nvdev)
1002 {
1003         struct netvsc_device_info *dev_info;
1004         struct bpf_prog *prog;
1005
1006         dev_info = kzalloc(sizeof(*dev_info), GFP_ATOMIC);
1007
1008         if (!dev_info)
1009                 return NULL;
1010
1011         if (nvdev) {
1012                 ASSERT_RTNL();
1013
1014                 dev_info->num_chn = nvdev->num_chn;
1015                 dev_info->send_sections = nvdev->send_section_cnt;
1016                 dev_info->send_section_size = nvdev->send_section_size;
1017                 dev_info->recv_sections = nvdev->recv_section_cnt;
1018                 dev_info->recv_section_size = nvdev->recv_section_size;
1019
1020                 memcpy(dev_info->rss_key, nvdev->extension->rss_key,
1021                        NETVSC_HASH_KEYLEN);
1022
1023                 prog = netvsc_xdp_get(nvdev);
1024                 if (prog) {
1025                         bpf_prog_inc(prog);
1026                         dev_info->bprog = prog;
1027                 }
1028         } else {
1029                 dev_info->num_chn = VRSS_CHANNEL_DEFAULT;
1030                 dev_info->send_sections = NETVSC_DEFAULT_TX;
1031                 dev_info->send_section_size = NETVSC_SEND_SECTION_SIZE;
1032                 dev_info->recv_sections = NETVSC_DEFAULT_RX;
1033                 dev_info->recv_section_size = NETVSC_RECV_SECTION_SIZE;
1034         }
1035
1036         return dev_info;
1037 }
1038
1039 /* Free struct netvsc_device_info */
1040 static void netvsc_devinfo_put(struct netvsc_device_info *dev_info)
1041 {
1042         if (dev_info->bprog) {
1043                 ASSERT_RTNL();
1044                 bpf_prog_put(dev_info->bprog);
1045         }
1046
1047         kfree(dev_info);
1048 }
1049
1050 static int netvsc_detach(struct net_device *ndev,
1051                          struct netvsc_device *nvdev)
1052 {
1053         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1054         struct hv_device *hdev = ndev_ctx->device_ctx;
1055         int ret;
1056
1057         /* Don't try continuing to try and setup sub channels */
1058         if (cancel_work_sync(&nvdev->subchan_work))
1059                 nvdev->num_chn = 1;
1060
1061         netvsc_xdp_set(ndev, NULL, NULL, nvdev);
1062
1063         /* If device was up (receiving) then shutdown */
1064         if (netif_running(ndev)) {
1065                 netvsc_tx_disable(nvdev, ndev);
1066
1067                 ret = rndis_filter_close(nvdev);
1068                 if (ret) {
1069                         netdev_err(ndev,
1070                                    "unable to close device (ret %d).\n", ret);
1071                         return ret;
1072                 }
1073
1074                 ret = netvsc_wait_until_empty(nvdev);
1075                 if (ret) {
1076                         netdev_err(ndev,
1077                                    "Ring buffer not empty after closing rndis\n");
1078                         return ret;
1079                 }
1080         }
1081
1082         netif_device_detach(ndev);
1083
1084         rndis_filter_device_remove(hdev, nvdev);
1085
1086         return 0;
1087 }
1088
1089 static int netvsc_attach(struct net_device *ndev,
1090                          struct netvsc_device_info *dev_info)
1091 {
1092         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1093         struct hv_device *hdev = ndev_ctx->device_ctx;
1094         struct netvsc_device *nvdev;
1095         struct rndis_device *rdev;
1096         struct bpf_prog *prog;
1097         int ret = 0;
1098
1099         nvdev = rndis_filter_device_add(hdev, dev_info);
1100         if (IS_ERR(nvdev))
1101                 return PTR_ERR(nvdev);
1102
1103         if (nvdev->num_chn > 1) {
1104                 ret = rndis_set_subchannel(ndev, nvdev, dev_info);
1105
1106                 /* if unavailable, just proceed with one queue */
1107                 if (ret) {
1108                         nvdev->max_chn = 1;
1109                         nvdev->num_chn = 1;
1110                 }
1111         }
1112
1113         prog = dev_info->bprog;
1114         if (prog) {
1115                 bpf_prog_inc(prog);
1116                 ret = netvsc_xdp_set(ndev, prog, NULL, nvdev);
1117                 if (ret) {
1118                         bpf_prog_put(prog);
1119                         goto err1;
1120                 }
1121         }
1122
1123         /* In any case device is now ready */
1124         nvdev->tx_disable = false;
1125         netif_device_attach(ndev);
1126
1127         /* Note: enable and attach happen when sub-channels setup */
1128         netif_carrier_off(ndev);
1129
1130         if (netif_running(ndev)) {
1131                 ret = rndis_filter_open(nvdev);
1132                 if (ret)
1133                         goto err2;
1134
1135                 rdev = nvdev->extension;
1136                 if (!rdev->link_state)
1137                         netif_carrier_on(ndev);
1138         }
1139
1140         return 0;
1141
1142 err2:
1143         netif_device_detach(ndev);
1144
1145 err1:
1146         rndis_filter_device_remove(hdev, nvdev);
1147
1148         return ret;
1149 }
1150
1151 static int netvsc_set_channels(struct net_device *net,
1152                                struct ethtool_channels *channels)
1153 {
1154         struct net_device_context *net_device_ctx = netdev_priv(net);
1155         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
1156         unsigned int orig, count = channels->combined_count;
1157         struct netvsc_device_info *device_info;
1158         int ret;
1159
1160         /* We do not support separate count for rx, tx, or other */
1161         if (count == 0 ||
1162             channels->rx_count || channels->tx_count || channels->other_count)
1163                 return -EINVAL;
1164
1165         if (!nvdev || nvdev->destroy)
1166                 return -ENODEV;
1167
1168         if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
1169                 return -EINVAL;
1170
1171         if (count > nvdev->max_chn)
1172                 return -EINVAL;
1173
1174         orig = nvdev->num_chn;
1175
1176         device_info = netvsc_devinfo_get(nvdev);
1177
1178         if (!device_info)
1179                 return -ENOMEM;
1180
1181         device_info->num_chn = count;
1182
1183         ret = netvsc_detach(net, nvdev);
1184         if (ret)
1185                 goto out;
1186
1187         ret = netvsc_attach(net, device_info);
1188         if (ret) {
1189                 device_info->num_chn = orig;
1190                 if (netvsc_attach(net, device_info))
1191                         netdev_err(net, "restoring channel setting failed\n");
1192         }
1193
1194 out:
1195         netvsc_devinfo_put(device_info);
1196         return ret;
1197 }
1198
1199 static void netvsc_init_settings(struct net_device *dev)
1200 {
1201         struct net_device_context *ndc = netdev_priv(dev);
1202
1203         ndc->l4_hash = HV_DEFAULT_L4HASH;
1204
1205         ndc->speed = SPEED_UNKNOWN;
1206         ndc->duplex = DUPLEX_FULL;
1207
1208         dev->features = NETIF_F_LRO;
1209 }
1210
1211 static int netvsc_get_link_ksettings(struct net_device *dev,
1212                                      struct ethtool_link_ksettings *cmd)
1213 {
1214         struct net_device_context *ndc = netdev_priv(dev);
1215         struct net_device *vf_netdev;
1216
1217         vf_netdev = rtnl_dereference(ndc->vf_netdev);
1218
1219         if (vf_netdev)
1220                 return __ethtool_get_link_ksettings(vf_netdev, cmd);
1221
1222         cmd->base.speed = ndc->speed;
1223         cmd->base.duplex = ndc->duplex;
1224         cmd->base.port = PORT_OTHER;
1225
1226         return 0;
1227 }
1228
1229 static int netvsc_set_link_ksettings(struct net_device *dev,
1230                                      const struct ethtool_link_ksettings *cmd)
1231 {
1232         struct net_device_context *ndc = netdev_priv(dev);
1233         struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1234
1235         if (vf_netdev) {
1236                 if (!vf_netdev->ethtool_ops->set_link_ksettings)
1237                         return -EOPNOTSUPP;
1238
1239                 return vf_netdev->ethtool_ops->set_link_ksettings(vf_netdev,
1240                                                                   cmd);
1241         }
1242
1243         return ethtool_virtdev_set_link_ksettings(dev, cmd,
1244                                                   &ndc->speed, &ndc->duplex);
1245 }
1246
1247 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
1248 {
1249         struct net_device_context *ndevctx = netdev_priv(ndev);
1250         struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1251         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1252         int orig_mtu = ndev->mtu;
1253         struct netvsc_device_info *device_info;
1254         int ret = 0;
1255
1256         if (!nvdev || nvdev->destroy)
1257                 return -ENODEV;
1258
1259         device_info = netvsc_devinfo_get(nvdev);
1260
1261         if (!device_info)
1262                 return -ENOMEM;
1263
1264         /* Change MTU of underlying VF netdev first. */
1265         if (vf_netdev) {
1266                 ret = dev_set_mtu(vf_netdev, mtu);
1267                 if (ret)
1268                         goto out;
1269         }
1270
1271         ret = netvsc_detach(ndev, nvdev);
1272         if (ret)
1273                 goto rollback_vf;
1274
1275         ndev->mtu = mtu;
1276
1277         ret = netvsc_attach(ndev, device_info);
1278         if (!ret)
1279                 goto out;
1280
1281         /* Attempt rollback to original MTU */
1282         ndev->mtu = orig_mtu;
1283
1284         if (netvsc_attach(ndev, device_info))
1285                 netdev_err(ndev, "restoring mtu failed\n");
1286 rollback_vf:
1287         if (vf_netdev)
1288                 dev_set_mtu(vf_netdev, orig_mtu);
1289
1290 out:
1291         netvsc_devinfo_put(device_info);
1292         return ret;
1293 }
1294
1295 static void netvsc_get_vf_stats(struct net_device *net,
1296                                 struct netvsc_vf_pcpu_stats *tot)
1297 {
1298         struct net_device_context *ndev_ctx = netdev_priv(net);
1299         int i;
1300
1301         memset(tot, 0, sizeof(*tot));
1302
1303         for_each_possible_cpu(i) {
1304                 const struct netvsc_vf_pcpu_stats *stats
1305                         = per_cpu_ptr(ndev_ctx->vf_stats, i);
1306                 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1307                 unsigned int start;
1308
1309                 do {
1310                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1311                         rx_packets = stats->rx_packets;
1312                         tx_packets = stats->tx_packets;
1313                         rx_bytes = stats->rx_bytes;
1314                         tx_bytes = stats->tx_bytes;
1315                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1316
1317                 tot->rx_packets += rx_packets;
1318                 tot->tx_packets += tx_packets;
1319                 tot->rx_bytes   += rx_bytes;
1320                 tot->tx_bytes   += tx_bytes;
1321                 tot->tx_dropped += stats->tx_dropped;
1322         }
1323 }
1324
1325 static void netvsc_get_pcpu_stats(struct net_device *net,
1326                                   struct netvsc_ethtool_pcpu_stats *pcpu_tot)
1327 {
1328         struct net_device_context *ndev_ctx = netdev_priv(net);
1329         struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
1330         int i;
1331
1332         /* fetch percpu stats of vf */
1333         for_each_possible_cpu(i) {
1334                 const struct netvsc_vf_pcpu_stats *stats =
1335                         per_cpu_ptr(ndev_ctx->vf_stats, i);
1336                 struct netvsc_ethtool_pcpu_stats *this_tot = &pcpu_tot[i];
1337                 unsigned int start;
1338
1339                 do {
1340                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1341                         this_tot->vf_rx_packets = stats->rx_packets;
1342                         this_tot->vf_tx_packets = stats->tx_packets;
1343                         this_tot->vf_rx_bytes = stats->rx_bytes;
1344                         this_tot->vf_tx_bytes = stats->tx_bytes;
1345                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1346                 this_tot->rx_packets = this_tot->vf_rx_packets;
1347                 this_tot->tx_packets = this_tot->vf_tx_packets;
1348                 this_tot->rx_bytes   = this_tot->vf_rx_bytes;
1349                 this_tot->tx_bytes   = this_tot->vf_tx_bytes;
1350         }
1351
1352         /* fetch percpu stats of netvsc */
1353         for (i = 0; i < nvdev->num_chn; i++) {
1354                 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1355                 const struct netvsc_stats *stats;
1356                 struct netvsc_ethtool_pcpu_stats *this_tot =
1357                         &pcpu_tot[nvchan->channel->target_cpu];
1358                 u64 packets, bytes;
1359                 unsigned int start;
1360
1361                 stats = &nvchan->tx_stats;
1362                 do {
1363                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1364                         packets = stats->packets;
1365                         bytes = stats->bytes;
1366                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1367
1368                 this_tot->tx_bytes      += bytes;
1369                 this_tot->tx_packets    += packets;
1370
1371                 stats = &nvchan->rx_stats;
1372                 do {
1373                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1374                         packets = stats->packets;
1375                         bytes = stats->bytes;
1376                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1377
1378                 this_tot->rx_bytes      += bytes;
1379                 this_tot->rx_packets    += packets;
1380         }
1381 }
1382
1383 static void netvsc_get_stats64(struct net_device *net,
1384                                struct rtnl_link_stats64 *t)
1385 {
1386         struct net_device_context *ndev_ctx = netdev_priv(net);
1387         struct netvsc_device *nvdev;
1388         struct netvsc_vf_pcpu_stats vf_tot;
1389         int i;
1390
1391         rcu_read_lock();
1392
1393         nvdev = rcu_dereference(ndev_ctx->nvdev);
1394         if (!nvdev)
1395                 goto out;
1396
1397         netdev_stats_to_stats64(t, &net->stats);
1398
1399         netvsc_get_vf_stats(net, &vf_tot);
1400         t->rx_packets += vf_tot.rx_packets;
1401         t->tx_packets += vf_tot.tx_packets;
1402         t->rx_bytes   += vf_tot.rx_bytes;
1403         t->tx_bytes   += vf_tot.tx_bytes;
1404         t->tx_dropped += vf_tot.tx_dropped;
1405
1406         for (i = 0; i < nvdev->num_chn; i++) {
1407                 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1408                 const struct netvsc_stats *stats;
1409                 u64 packets, bytes, multicast;
1410                 unsigned int start;
1411
1412                 stats = &nvchan->tx_stats;
1413                 do {
1414                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1415                         packets = stats->packets;
1416                         bytes = stats->bytes;
1417                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1418
1419                 t->tx_bytes     += bytes;
1420                 t->tx_packets   += packets;
1421
1422                 stats = &nvchan->rx_stats;
1423                 do {
1424                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1425                         packets = stats->packets;
1426                         bytes = stats->bytes;
1427                         multicast = stats->multicast + stats->broadcast;
1428                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1429
1430                 t->rx_bytes     += bytes;
1431                 t->rx_packets   += packets;
1432                 t->multicast    += multicast;
1433         }
1434 out:
1435         rcu_read_unlock();
1436 }
1437
1438 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
1439 {
1440         struct net_device_context *ndc = netdev_priv(ndev);
1441         struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1442         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1443         struct sockaddr *addr = p;
1444         int err;
1445
1446         err = eth_prepare_mac_addr_change(ndev, p);
1447         if (err)
1448                 return err;
1449
1450         if (!nvdev)
1451                 return -ENODEV;
1452
1453         if (vf_netdev) {
1454                 err = dev_set_mac_address(vf_netdev, addr, NULL);
1455                 if (err)
1456                         return err;
1457         }
1458
1459         err = rndis_filter_set_device_mac(nvdev, addr->sa_data);
1460         if (!err) {
1461                 eth_commit_mac_addr_change(ndev, p);
1462         } else if (vf_netdev) {
1463                 /* rollback change on VF */
1464                 memcpy(addr->sa_data, ndev->dev_addr, ETH_ALEN);
1465                 dev_set_mac_address(vf_netdev, addr, NULL);
1466         }
1467
1468         return err;
1469 }
1470
1471 static const struct {
1472         char name[ETH_GSTRING_LEN];
1473         u16 offset;
1474 } netvsc_stats[] = {
1475         { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
1476         { "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
1477         { "tx_no_space",  offsetof(struct netvsc_ethtool_stats, tx_no_space) },
1478         { "tx_too_big",   offsetof(struct netvsc_ethtool_stats, tx_too_big) },
1479         { "tx_busy",      offsetof(struct netvsc_ethtool_stats, tx_busy) },
1480         { "tx_send_full", offsetof(struct netvsc_ethtool_stats, tx_send_full) },
1481         { "rx_comp_busy", offsetof(struct netvsc_ethtool_stats, rx_comp_busy) },
1482         { "rx_no_memory", offsetof(struct netvsc_ethtool_stats, rx_no_memory) },
1483         { "stop_queue", offsetof(struct netvsc_ethtool_stats, stop_queue) },
1484         { "wake_queue", offsetof(struct netvsc_ethtool_stats, wake_queue) },
1485         { "vlan_error", offsetof(struct netvsc_ethtool_stats, vlan_error) },
1486 }, pcpu_stats[] = {
1487         { "cpu%u_rx_packets",
1488                 offsetof(struct netvsc_ethtool_pcpu_stats, rx_packets) },
1489         { "cpu%u_rx_bytes",
1490                 offsetof(struct netvsc_ethtool_pcpu_stats, rx_bytes) },
1491         { "cpu%u_tx_packets",
1492                 offsetof(struct netvsc_ethtool_pcpu_stats, tx_packets) },
1493         { "cpu%u_tx_bytes",
1494                 offsetof(struct netvsc_ethtool_pcpu_stats, tx_bytes) },
1495         { "cpu%u_vf_rx_packets",
1496                 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_packets) },
1497         { "cpu%u_vf_rx_bytes",
1498                 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_bytes) },
1499         { "cpu%u_vf_tx_packets",
1500                 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_packets) },
1501         { "cpu%u_vf_tx_bytes",
1502                 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_bytes) },
1503 }, vf_stats[] = {
1504         { "vf_rx_packets", offsetof(struct netvsc_vf_pcpu_stats, rx_packets) },
1505         { "vf_rx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, rx_bytes) },
1506         { "vf_tx_packets", offsetof(struct netvsc_vf_pcpu_stats, tx_packets) },
1507         { "vf_tx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, tx_bytes) },
1508         { "vf_tx_dropped", offsetof(struct netvsc_vf_pcpu_stats, tx_dropped) },
1509 };
1510
1511 #define NETVSC_GLOBAL_STATS_LEN ARRAY_SIZE(netvsc_stats)
1512 #define NETVSC_VF_STATS_LEN     ARRAY_SIZE(vf_stats)
1513
1514 /* statistics per queue (rx/tx packets/bytes) */
1515 #define NETVSC_PCPU_STATS_LEN (num_present_cpus() * ARRAY_SIZE(pcpu_stats))
1516
1517 /* 5 statistics per queue (rx/tx packets/bytes, rx xdp_drop) */
1518 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 5)
1519
1520 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
1521 {
1522         struct net_device_context *ndc = netdev_priv(dev);
1523         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1524
1525         if (!nvdev)
1526                 return -ENODEV;
1527
1528         switch (string_set) {
1529         case ETH_SS_STATS:
1530                 return NETVSC_GLOBAL_STATS_LEN
1531                         + NETVSC_VF_STATS_LEN
1532                         + NETVSC_QUEUE_STATS_LEN(nvdev)
1533                         + NETVSC_PCPU_STATS_LEN;
1534         default:
1535                 return -EINVAL;
1536         }
1537 }
1538
1539 static void netvsc_get_ethtool_stats(struct net_device *dev,
1540                                      struct ethtool_stats *stats, u64 *data)
1541 {
1542         struct net_device_context *ndc = netdev_priv(dev);
1543         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1544         const void *nds = &ndc->eth_stats;
1545         const struct netvsc_stats *qstats;
1546         struct netvsc_vf_pcpu_stats sum;
1547         struct netvsc_ethtool_pcpu_stats *pcpu_sum;
1548         unsigned int start;
1549         u64 packets, bytes;
1550         u64 xdp_drop;
1551         int i, j, cpu;
1552
1553         if (!nvdev)
1554                 return;
1555
1556         for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
1557                 data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1558
1559         netvsc_get_vf_stats(dev, &sum);
1560         for (j = 0; j < NETVSC_VF_STATS_LEN; j++)
1561                 data[i++] = *(u64 *)((void *)&sum + vf_stats[j].offset);
1562
1563         for (j = 0; j < nvdev->num_chn; j++) {
1564                 qstats = &nvdev->chan_table[j].tx_stats;
1565
1566                 do {
1567                         start = u64_stats_fetch_begin_irq(&qstats->syncp);
1568                         packets = qstats->packets;
1569                         bytes = qstats->bytes;
1570                 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1571                 data[i++] = packets;
1572                 data[i++] = bytes;
1573
1574                 qstats = &nvdev->chan_table[j].rx_stats;
1575                 do {
1576                         start = u64_stats_fetch_begin_irq(&qstats->syncp);
1577                         packets = qstats->packets;
1578                         bytes = qstats->bytes;
1579                         xdp_drop = qstats->xdp_drop;
1580                 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1581                 data[i++] = packets;
1582                 data[i++] = bytes;
1583                 data[i++] = xdp_drop;
1584         }
1585
1586         pcpu_sum = kvmalloc_array(num_possible_cpus(),
1587                                   sizeof(struct netvsc_ethtool_pcpu_stats),
1588                                   GFP_KERNEL);
1589         netvsc_get_pcpu_stats(dev, pcpu_sum);
1590         for_each_present_cpu(cpu) {
1591                 struct netvsc_ethtool_pcpu_stats *this_sum = &pcpu_sum[cpu];
1592
1593                 for (j = 0; j < ARRAY_SIZE(pcpu_stats); j++)
1594                         data[i++] = *(u64 *)((void *)this_sum
1595                                              + pcpu_stats[j].offset);
1596         }
1597         kvfree(pcpu_sum);
1598 }
1599
1600 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1601 {
1602         struct net_device_context *ndc = netdev_priv(dev);
1603         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1604         u8 *p = data;
1605         int i, cpu;
1606
1607         if (!nvdev)
1608                 return;
1609
1610         switch (stringset) {
1611         case ETH_SS_STATS:
1612                 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++)
1613                         ethtool_sprintf(&p, netvsc_stats[i].name);
1614
1615                 for (i = 0; i < ARRAY_SIZE(vf_stats); i++)
1616                         ethtool_sprintf(&p, vf_stats[i].name);
1617
1618                 for (i = 0; i < nvdev->num_chn; i++) {
1619                         ethtool_sprintf(&p, "tx_queue_%u_packets", i);
1620                         ethtool_sprintf(&p, "tx_queue_%u_bytes", i);
1621                         ethtool_sprintf(&p, "rx_queue_%u_packets", i);
1622                         ethtool_sprintf(&p, "rx_queue_%u_bytes", i);
1623                         ethtool_sprintf(&p, "rx_queue_%u_xdp_drop", i);
1624                 }
1625
1626                 for_each_present_cpu(cpu) {
1627                         for (i = 0; i < ARRAY_SIZE(pcpu_stats); i++)
1628                                 ethtool_sprintf(&p, pcpu_stats[i].name, cpu);
1629                 }
1630
1631                 break;
1632         }
1633 }
1634
1635 static int
1636 netvsc_get_rss_hash_opts(struct net_device_context *ndc,
1637                          struct ethtool_rxnfc *info)
1638 {
1639         const u32 l4_flag = RXH_L4_B_0_1 | RXH_L4_B_2_3;
1640
1641         info->data = RXH_IP_SRC | RXH_IP_DST;
1642
1643         switch (info->flow_type) {
1644         case TCP_V4_FLOW:
1645                 if (ndc->l4_hash & HV_TCP4_L4HASH)
1646                         info->data |= l4_flag;
1647
1648                 break;
1649
1650         case TCP_V6_FLOW:
1651                 if (ndc->l4_hash & HV_TCP6_L4HASH)
1652                         info->data |= l4_flag;
1653
1654                 break;
1655
1656         case UDP_V4_FLOW:
1657                 if (ndc->l4_hash & HV_UDP4_L4HASH)
1658                         info->data |= l4_flag;
1659
1660                 break;
1661
1662         case UDP_V6_FLOW:
1663                 if (ndc->l4_hash & HV_UDP6_L4HASH)
1664                         info->data |= l4_flag;
1665
1666                 break;
1667
1668         case IPV4_FLOW:
1669         case IPV6_FLOW:
1670                 break;
1671         default:
1672                 info->data = 0;
1673                 break;
1674         }
1675
1676         return 0;
1677 }
1678
1679 static int
1680 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1681                  u32 *rules)
1682 {
1683         struct net_device_context *ndc = netdev_priv(dev);
1684         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1685
1686         if (!nvdev)
1687                 return -ENODEV;
1688
1689         switch (info->cmd) {
1690         case ETHTOOL_GRXRINGS:
1691                 info->data = nvdev->num_chn;
1692                 return 0;
1693
1694         case ETHTOOL_GRXFH:
1695                 return netvsc_get_rss_hash_opts(ndc, info);
1696         }
1697         return -EOPNOTSUPP;
1698 }
1699
1700 static int netvsc_set_rss_hash_opts(struct net_device_context *ndc,
1701                                     struct ethtool_rxnfc *info)
1702 {
1703         if (info->data == (RXH_IP_SRC | RXH_IP_DST |
1704                            RXH_L4_B_0_1 | RXH_L4_B_2_3)) {
1705                 switch (info->flow_type) {
1706                 case TCP_V4_FLOW:
1707                         ndc->l4_hash |= HV_TCP4_L4HASH;
1708                         break;
1709
1710                 case TCP_V6_FLOW:
1711                         ndc->l4_hash |= HV_TCP6_L4HASH;
1712                         break;
1713
1714                 case UDP_V4_FLOW:
1715                         ndc->l4_hash |= HV_UDP4_L4HASH;
1716                         break;
1717
1718                 case UDP_V6_FLOW:
1719                         ndc->l4_hash |= HV_UDP6_L4HASH;
1720                         break;
1721
1722                 default:
1723                         return -EOPNOTSUPP;
1724                 }
1725
1726                 return 0;
1727         }
1728
1729         if (info->data == (RXH_IP_SRC | RXH_IP_DST)) {
1730                 switch (info->flow_type) {
1731                 case TCP_V4_FLOW:
1732                         ndc->l4_hash &= ~HV_TCP4_L4HASH;
1733                         break;
1734
1735                 case TCP_V6_FLOW:
1736                         ndc->l4_hash &= ~HV_TCP6_L4HASH;
1737                         break;
1738
1739                 case UDP_V4_FLOW:
1740                         ndc->l4_hash &= ~HV_UDP4_L4HASH;
1741                         break;
1742
1743                 case UDP_V6_FLOW:
1744                         ndc->l4_hash &= ~HV_UDP6_L4HASH;
1745                         break;
1746
1747                 default:
1748                         return -EOPNOTSUPP;
1749                 }
1750
1751                 return 0;
1752         }
1753
1754         return -EOPNOTSUPP;
1755 }
1756
1757 static int
1758 netvsc_set_rxnfc(struct net_device *ndev, struct ethtool_rxnfc *info)
1759 {
1760         struct net_device_context *ndc = netdev_priv(ndev);
1761
1762         if (info->cmd == ETHTOOL_SRXFH)
1763                 return netvsc_set_rss_hash_opts(ndc, info);
1764
1765         return -EOPNOTSUPP;
1766 }
1767
1768 static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1769 {
1770         return NETVSC_HASH_KEYLEN;
1771 }
1772
1773 static u32 netvsc_rss_indir_size(struct net_device *dev)
1774 {
1775         return ITAB_NUM;
1776 }
1777
1778 static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key,
1779                            u8 *hfunc)
1780 {
1781         struct net_device_context *ndc = netdev_priv(dev);
1782         struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1783         struct rndis_device *rndis_dev;
1784         int i;
1785
1786         if (!ndev)
1787                 return -ENODEV;
1788
1789         if (hfunc)
1790                 *hfunc = ETH_RSS_HASH_TOP;      /* Toeplitz */
1791
1792         rndis_dev = ndev->extension;
1793         if (indir) {
1794                 for (i = 0; i < ITAB_NUM; i++)
1795                         indir[i] = ndc->rx_table[i];
1796         }
1797
1798         if (key)
1799                 memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1800
1801         return 0;
1802 }
1803
1804 static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
1805                            const u8 *key, const u8 hfunc)
1806 {
1807         struct net_device_context *ndc = netdev_priv(dev);
1808         struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1809         struct rndis_device *rndis_dev;
1810         int i;
1811
1812         if (!ndev)
1813                 return -ENODEV;
1814
1815         if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
1816                 return -EOPNOTSUPP;
1817
1818         rndis_dev = ndev->extension;
1819         if (indir) {
1820                 for (i = 0; i < ITAB_NUM; i++)
1821                         if (indir[i] >= ndev->num_chn)
1822                                 return -EINVAL;
1823
1824                 for (i = 0; i < ITAB_NUM; i++)
1825                         ndc->rx_table[i] = indir[i];
1826         }
1827
1828         if (!key) {
1829                 if (!indir)
1830                         return 0;
1831
1832                 key = rndis_dev->rss_key;
1833         }
1834
1835         return rndis_filter_set_rss_param(rndis_dev, key);
1836 }
1837
1838 /* Hyper-V RNDIS protocol does not have ring in the HW sense.
1839  * It does have pre-allocated receive area which is divided into sections.
1840  */
1841 static void __netvsc_get_ringparam(struct netvsc_device *nvdev,
1842                                    struct ethtool_ringparam *ring)
1843 {
1844         u32 max_buf_size;
1845
1846         ring->rx_pending = nvdev->recv_section_cnt;
1847         ring->tx_pending = nvdev->send_section_cnt;
1848
1849         if (nvdev->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
1850                 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
1851         else
1852                 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
1853
1854         ring->rx_max_pending = max_buf_size / nvdev->recv_section_size;
1855         ring->tx_max_pending = NETVSC_SEND_BUFFER_SIZE
1856                 / nvdev->send_section_size;
1857 }
1858
1859 static void netvsc_get_ringparam(struct net_device *ndev,
1860                                  struct ethtool_ringparam *ring)
1861 {
1862         struct net_device_context *ndevctx = netdev_priv(ndev);
1863         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1864
1865         if (!nvdev)
1866                 return;
1867
1868         __netvsc_get_ringparam(nvdev, ring);
1869 }
1870
1871 static int netvsc_set_ringparam(struct net_device *ndev,
1872                                 struct ethtool_ringparam *ring)
1873 {
1874         struct net_device_context *ndevctx = netdev_priv(ndev);
1875         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1876         struct netvsc_device_info *device_info;
1877         struct ethtool_ringparam orig;
1878         u32 new_tx, new_rx;
1879         int ret = 0;
1880
1881         if (!nvdev || nvdev->destroy)
1882                 return -ENODEV;
1883
1884         memset(&orig, 0, sizeof(orig));
1885         __netvsc_get_ringparam(nvdev, &orig);
1886
1887         new_tx = clamp_t(u32, ring->tx_pending,
1888                          NETVSC_MIN_TX_SECTIONS, orig.tx_max_pending);
1889         new_rx = clamp_t(u32, ring->rx_pending,
1890                          NETVSC_MIN_RX_SECTIONS, orig.rx_max_pending);
1891
1892         if (new_tx == orig.tx_pending &&
1893             new_rx == orig.rx_pending)
1894                 return 0;        /* no change */
1895
1896         device_info = netvsc_devinfo_get(nvdev);
1897
1898         if (!device_info)
1899                 return -ENOMEM;
1900
1901         device_info->send_sections = new_tx;
1902         device_info->recv_sections = new_rx;
1903
1904         ret = netvsc_detach(ndev, nvdev);
1905         if (ret)
1906                 goto out;
1907
1908         ret = netvsc_attach(ndev, device_info);
1909         if (ret) {
1910                 device_info->send_sections = orig.tx_pending;
1911                 device_info->recv_sections = orig.rx_pending;
1912
1913                 if (netvsc_attach(ndev, device_info))
1914                         netdev_err(ndev, "restoring ringparam failed");
1915         }
1916
1917 out:
1918         netvsc_devinfo_put(device_info);
1919         return ret;
1920 }
1921
1922 static netdev_features_t netvsc_fix_features(struct net_device *ndev,
1923                                              netdev_features_t features)
1924 {
1925         struct net_device_context *ndevctx = netdev_priv(ndev);
1926         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1927
1928         if (!nvdev || nvdev->destroy)
1929                 return features;
1930
1931         if ((features & NETIF_F_LRO) && netvsc_xdp_get(nvdev)) {
1932                 features ^= NETIF_F_LRO;
1933                 netdev_info(ndev, "Skip LRO - unsupported with XDP\n");
1934         }
1935
1936         return features;
1937 }
1938
1939 static int netvsc_set_features(struct net_device *ndev,
1940                                netdev_features_t features)
1941 {
1942         netdev_features_t change = features ^ ndev->features;
1943         struct net_device_context *ndevctx = netdev_priv(ndev);
1944         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1945         struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1946         struct ndis_offload_params offloads;
1947         int ret = 0;
1948
1949         if (!nvdev || nvdev->destroy)
1950                 return -ENODEV;
1951
1952         if (!(change & NETIF_F_LRO))
1953                 goto syncvf;
1954
1955         memset(&offloads, 0, sizeof(struct ndis_offload_params));
1956
1957         if (features & NETIF_F_LRO) {
1958                 offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1959                 offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1960         } else {
1961                 offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1962                 offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1963         }
1964
1965         ret = rndis_filter_set_offload_params(ndev, nvdev, &offloads);
1966
1967         if (ret) {
1968                 features ^= NETIF_F_LRO;
1969                 ndev->features = features;
1970         }
1971
1972 syncvf:
1973         if (!vf_netdev)
1974                 return ret;
1975
1976         vf_netdev->wanted_features = features;
1977         netdev_update_features(vf_netdev);
1978
1979         return ret;
1980 }
1981
1982 static int netvsc_get_regs_len(struct net_device *netdev)
1983 {
1984         return VRSS_SEND_TAB_SIZE * sizeof(u32);
1985 }
1986
1987 static void netvsc_get_regs(struct net_device *netdev,
1988                             struct ethtool_regs *regs, void *p)
1989 {
1990         struct net_device_context *ndc = netdev_priv(netdev);
1991         u32 *regs_buff = p;
1992
1993         /* increase the version, if buffer format is changed. */
1994         regs->version = 1;
1995
1996         memcpy(regs_buff, ndc->tx_table, VRSS_SEND_TAB_SIZE * sizeof(u32));
1997 }
1998
1999 static u32 netvsc_get_msglevel(struct net_device *ndev)
2000 {
2001         struct net_device_context *ndev_ctx = netdev_priv(ndev);
2002
2003         return ndev_ctx->msg_enable;
2004 }
2005
2006 static void netvsc_set_msglevel(struct net_device *ndev, u32 val)
2007 {
2008         struct net_device_context *ndev_ctx = netdev_priv(ndev);
2009
2010         ndev_ctx->msg_enable = val;
2011 }
2012
2013 static const struct ethtool_ops ethtool_ops = {
2014         .get_drvinfo    = netvsc_get_drvinfo,
2015         .get_regs_len   = netvsc_get_regs_len,
2016         .get_regs       = netvsc_get_regs,
2017         .get_msglevel   = netvsc_get_msglevel,
2018         .set_msglevel   = netvsc_set_msglevel,
2019         .get_link       = ethtool_op_get_link,
2020         .get_ethtool_stats = netvsc_get_ethtool_stats,
2021         .get_sset_count = netvsc_get_sset_count,
2022         .get_strings    = netvsc_get_strings,
2023         .get_channels   = netvsc_get_channels,
2024         .set_channels   = netvsc_set_channels,
2025         .get_ts_info    = ethtool_op_get_ts_info,
2026         .get_rxnfc      = netvsc_get_rxnfc,
2027         .set_rxnfc      = netvsc_set_rxnfc,
2028         .get_rxfh_key_size = netvsc_get_rxfh_key_size,
2029         .get_rxfh_indir_size = netvsc_rss_indir_size,
2030         .get_rxfh       = netvsc_get_rxfh,
2031         .set_rxfh       = netvsc_set_rxfh,
2032         .get_link_ksettings = netvsc_get_link_ksettings,
2033         .set_link_ksettings = netvsc_set_link_ksettings,
2034         .get_ringparam  = netvsc_get_ringparam,
2035         .set_ringparam  = netvsc_set_ringparam,
2036 };
2037
2038 static const struct net_device_ops device_ops = {
2039         .ndo_open =                     netvsc_open,
2040         .ndo_stop =                     netvsc_close,
2041         .ndo_start_xmit =               netvsc_start_xmit,
2042         .ndo_change_rx_flags =          netvsc_change_rx_flags,
2043         .ndo_set_rx_mode =              netvsc_set_rx_mode,
2044         .ndo_fix_features =             netvsc_fix_features,
2045         .ndo_set_features =             netvsc_set_features,
2046         .ndo_change_mtu =               netvsc_change_mtu,
2047         .ndo_validate_addr =            eth_validate_addr,
2048         .ndo_set_mac_address =          netvsc_set_mac_addr,
2049         .ndo_select_queue =             netvsc_select_queue,
2050         .ndo_get_stats64 =              netvsc_get_stats64,
2051         .ndo_bpf =                      netvsc_bpf,
2052 };
2053
2054 /*
2055  * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
2056  * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
2057  * present send GARP packet to network peers with netif_notify_peers().
2058  */
2059 static void netvsc_link_change(struct work_struct *w)
2060 {
2061         struct net_device_context *ndev_ctx =
2062                 container_of(w, struct net_device_context, dwork.work);
2063         struct hv_device *device_obj = ndev_ctx->device_ctx;
2064         struct net_device *net = hv_get_drvdata(device_obj);
2065         unsigned long flags, next_reconfig, delay;
2066         struct netvsc_reconfig *event = NULL;
2067         struct netvsc_device *net_device;
2068         struct rndis_device *rdev;
2069         bool reschedule = false;
2070
2071         /* if changes are happening, comeback later */
2072         if (!rtnl_trylock()) {
2073                 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
2074                 return;
2075         }
2076
2077         net_device = rtnl_dereference(ndev_ctx->nvdev);
2078         if (!net_device)
2079                 goto out_unlock;
2080
2081         rdev = net_device->extension;
2082
2083         next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
2084         if (time_is_after_jiffies(next_reconfig)) {
2085                 /* link_watch only sends one notification with current state
2086                  * per second, avoid doing reconfig more frequently. Handle
2087                  * wrap around.
2088                  */
2089                 delay = next_reconfig - jiffies;
2090                 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
2091                 schedule_delayed_work(&ndev_ctx->dwork, delay);
2092                 goto out_unlock;
2093         }
2094         ndev_ctx->last_reconfig = jiffies;
2095
2096         spin_lock_irqsave(&ndev_ctx->lock, flags);
2097         if (!list_empty(&ndev_ctx->reconfig_events)) {
2098                 event = list_first_entry(&ndev_ctx->reconfig_events,
2099                                          struct netvsc_reconfig, list);
2100                 list_del(&event->list);
2101                 reschedule = !list_empty(&ndev_ctx->reconfig_events);
2102         }
2103         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
2104
2105         if (!event)
2106                 goto out_unlock;
2107
2108         switch (event->event) {
2109                 /* Only the following events are possible due to the check in
2110                  * netvsc_linkstatus_callback()
2111                  */
2112         case RNDIS_STATUS_MEDIA_CONNECT:
2113                 if (rdev->link_state) {
2114                         rdev->link_state = false;
2115                         netif_carrier_on(net);
2116                         netvsc_tx_enable(net_device, net);
2117                 } else {
2118                         __netdev_notify_peers(net);
2119                 }
2120                 kfree(event);
2121                 break;
2122         case RNDIS_STATUS_MEDIA_DISCONNECT:
2123                 if (!rdev->link_state) {
2124                         rdev->link_state = true;
2125                         netif_carrier_off(net);
2126                         netvsc_tx_disable(net_device, net);
2127                 }
2128                 kfree(event);
2129                 break;
2130         case RNDIS_STATUS_NETWORK_CHANGE:
2131                 /* Only makes sense if carrier is present */
2132                 if (!rdev->link_state) {
2133                         rdev->link_state = true;
2134                         netif_carrier_off(net);
2135                         netvsc_tx_disable(net_device, net);
2136                         event->event = RNDIS_STATUS_MEDIA_CONNECT;
2137                         spin_lock_irqsave(&ndev_ctx->lock, flags);
2138                         list_add(&event->list, &ndev_ctx->reconfig_events);
2139                         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
2140                         reschedule = true;
2141                 }
2142                 break;
2143         }
2144
2145         rtnl_unlock();
2146
2147         /* link_watch only sends one notification with current state per
2148          * second, handle next reconfig event in 2 seconds.
2149          */
2150         if (reschedule)
2151                 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
2152
2153         return;
2154
2155 out_unlock:
2156         rtnl_unlock();
2157 }
2158
2159 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
2160 {
2161         struct net_device_context *net_device_ctx;
2162         struct net_device *dev;
2163
2164         dev = netdev_master_upper_dev_get(vf_netdev);
2165         if (!dev || dev->netdev_ops != &device_ops)
2166                 return NULL;    /* not a netvsc device */
2167
2168         net_device_ctx = netdev_priv(dev);
2169         if (!rtnl_dereference(net_device_ctx->nvdev))
2170                 return NULL;    /* device is removed */
2171
2172         return dev;
2173 }
2174
2175 /* Called when VF is injecting data into network stack.
2176  * Change the associated network device from VF to netvsc.
2177  * note: already called with rcu_read_lock
2178  */
2179 static rx_handler_result_t netvsc_vf_handle_frame(struct sk_buff **pskb)
2180 {
2181         struct sk_buff *skb = *pskb;
2182         struct net_device *ndev = rcu_dereference(skb->dev->rx_handler_data);
2183         struct net_device_context *ndev_ctx = netdev_priv(ndev);
2184         struct netvsc_vf_pcpu_stats *pcpu_stats
2185                  = this_cpu_ptr(ndev_ctx->vf_stats);
2186
2187         skb = skb_share_check(skb, GFP_ATOMIC);
2188         if (unlikely(!skb))
2189                 return RX_HANDLER_CONSUMED;
2190
2191         *pskb = skb;
2192
2193         skb->dev = ndev;
2194
2195         u64_stats_update_begin(&pcpu_stats->syncp);
2196         pcpu_stats->rx_packets++;
2197         pcpu_stats->rx_bytes += skb->len;
2198         u64_stats_update_end(&pcpu_stats->syncp);
2199
2200         return RX_HANDLER_ANOTHER;
2201 }
2202
2203 static int netvsc_vf_join(struct net_device *vf_netdev,
2204                           struct net_device *ndev)
2205 {
2206         struct net_device_context *ndev_ctx = netdev_priv(ndev);
2207         int ret;
2208
2209         ret = netdev_rx_handler_register(vf_netdev,
2210                                          netvsc_vf_handle_frame, ndev);
2211         if (ret != 0) {
2212                 netdev_err(vf_netdev,
2213                            "can not register netvsc VF receive handler (err = %d)\n",
2214                            ret);
2215                 goto rx_handler_failed;
2216         }
2217
2218         ret = netdev_master_upper_dev_link(vf_netdev, ndev,
2219                                            NULL, NULL, NULL);
2220         if (ret != 0) {
2221                 netdev_err(vf_netdev,
2222                            "can not set master device %s (err = %d)\n",
2223                            ndev->name, ret);
2224                 goto upper_link_failed;
2225         }
2226
2227         /* set slave flag before open to prevent IPv6 addrconf */
2228         vf_netdev->flags |= IFF_SLAVE;
2229
2230         schedule_delayed_work(&ndev_ctx->vf_takeover, VF_TAKEOVER_INT);
2231
2232         call_netdevice_notifiers(NETDEV_JOIN, vf_netdev);
2233
2234         netdev_info(vf_netdev, "joined to %s\n", ndev->name);
2235         return 0;
2236
2237 upper_link_failed:
2238         netdev_rx_handler_unregister(vf_netdev);
2239 rx_handler_failed:
2240         return ret;
2241 }
2242
2243 static void __netvsc_vf_setup(struct net_device *ndev,
2244                               struct net_device *vf_netdev)
2245 {
2246         int ret;
2247
2248         /* Align MTU of VF with master */
2249         ret = dev_set_mtu(vf_netdev, ndev->mtu);
2250         if (ret)
2251                 netdev_warn(vf_netdev,
2252                             "unable to change mtu to %u\n", ndev->mtu);
2253
2254         /* set multicast etc flags on VF */
2255         dev_change_flags(vf_netdev, ndev->flags | IFF_SLAVE, NULL);
2256
2257         /* sync address list from ndev to VF */
2258         netif_addr_lock_bh(ndev);
2259         dev_uc_sync(vf_netdev, ndev);
2260         dev_mc_sync(vf_netdev, ndev);
2261         netif_addr_unlock_bh(ndev);
2262
2263         if (netif_running(ndev)) {
2264                 ret = dev_open(vf_netdev, NULL);
2265                 if (ret)
2266                         netdev_warn(vf_netdev,
2267                                     "unable to open: %d\n", ret);
2268         }
2269 }
2270
2271 /* Setup VF as slave of the synthetic device.
2272  * Runs in workqueue to avoid recursion in netlink callbacks.
2273  */
2274 static void netvsc_vf_setup(struct work_struct *w)
2275 {
2276         struct net_device_context *ndev_ctx
2277                 = container_of(w, struct net_device_context, vf_takeover.work);
2278         struct net_device *ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2279         struct net_device *vf_netdev;
2280
2281         if (!rtnl_trylock()) {
2282                 schedule_delayed_work(&ndev_ctx->vf_takeover, 0);
2283                 return;
2284         }
2285
2286         vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2287         if (vf_netdev)
2288                 __netvsc_vf_setup(ndev, vf_netdev);
2289
2290         rtnl_unlock();
2291 }
2292
2293 /* Find netvsc by VF serial number.
2294  * The PCI hyperv controller records the serial number as the slot kobj name.
2295  */
2296 static struct net_device *get_netvsc_byslot(const struct net_device *vf_netdev)
2297 {
2298         struct device *parent = vf_netdev->dev.parent;
2299         struct net_device_context *ndev_ctx;
2300         struct net_device *ndev;
2301         struct pci_dev *pdev;
2302         u32 serial;
2303
2304         if (!parent || !dev_is_pci(parent))
2305                 return NULL; /* not a PCI device */
2306
2307         pdev = to_pci_dev(parent);
2308         if (!pdev->slot) {
2309                 netdev_notice(vf_netdev, "no PCI slot information\n");
2310                 return NULL;
2311         }
2312
2313         if (kstrtou32(pci_slot_name(pdev->slot), 10, &serial)) {
2314                 netdev_notice(vf_netdev, "Invalid vf serial:%s\n",
2315                               pci_slot_name(pdev->slot));
2316                 return NULL;
2317         }
2318
2319         list_for_each_entry(ndev_ctx, &netvsc_dev_list, list) {
2320                 if (!ndev_ctx->vf_alloc)
2321                         continue;
2322
2323                 if (ndev_ctx->vf_serial != serial)
2324                         continue;
2325
2326                 ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2327                 if (ndev->addr_len != vf_netdev->addr_len ||
2328                     memcmp(ndev->perm_addr, vf_netdev->perm_addr,
2329                            ndev->addr_len) != 0)
2330                         continue;
2331
2332                 return ndev;
2333
2334         }
2335
2336         netdev_notice(vf_netdev,
2337                       "no netdev found for vf serial:%u\n", serial);
2338         return NULL;
2339 }
2340
2341 static int netvsc_register_vf(struct net_device *vf_netdev)
2342 {
2343         struct net_device_context *net_device_ctx;
2344         struct netvsc_device *netvsc_dev;
2345         struct bpf_prog *prog;
2346         struct net_device *ndev;
2347         int ret;
2348
2349         if (vf_netdev->addr_len != ETH_ALEN)
2350                 return NOTIFY_DONE;
2351
2352         ndev = get_netvsc_byslot(vf_netdev);
2353         if (!ndev)
2354                 return NOTIFY_DONE;
2355
2356         net_device_ctx = netdev_priv(ndev);
2357         netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2358         if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
2359                 return NOTIFY_DONE;
2360
2361         /* if synthetic interface is a different namespace,
2362          * then move the VF to that namespace; join will be
2363          * done again in that context.
2364          */
2365         if (!net_eq(dev_net(ndev), dev_net(vf_netdev))) {
2366                 ret = dev_change_net_namespace(vf_netdev,
2367                                                dev_net(ndev), "eth%d");
2368                 if (ret)
2369                         netdev_err(vf_netdev,
2370                                    "could not move to same namespace as %s: %d\n",
2371                                    ndev->name, ret);
2372                 else
2373                         netdev_info(vf_netdev,
2374                                     "VF moved to namespace with: %s\n",
2375                                     ndev->name);
2376                 return NOTIFY_DONE;
2377         }
2378
2379         netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
2380
2381         if (netvsc_vf_join(vf_netdev, ndev) != 0)
2382                 return NOTIFY_DONE;
2383
2384         dev_hold(vf_netdev);
2385         rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
2386
2387         if (ndev->needed_headroom < vf_netdev->needed_headroom)
2388                 ndev->needed_headroom = vf_netdev->needed_headroom;
2389
2390         vf_netdev->wanted_features = ndev->features;
2391         netdev_update_features(vf_netdev);
2392
2393         prog = netvsc_xdp_get(netvsc_dev);
2394         netvsc_vf_setxdp(vf_netdev, prog);
2395
2396         return NOTIFY_OK;
2397 }
2398
2399 /* Change the data path when VF UP/DOWN/CHANGE are detected.
2400  *
2401  * Typically a UP or DOWN event is followed by a CHANGE event, so
2402  * net_device_ctx->data_path_is_vf is used to cache the current data path
2403  * to avoid the duplicate call of netvsc_switch_datapath() and the duplicate
2404  * message.
2405  *
2406  * During hibernation, if a VF NIC driver (e.g. mlx5) preserves the network
2407  * interface, there is only the CHANGE event and no UP or DOWN event.
2408  */
2409 static int netvsc_vf_changed(struct net_device *vf_netdev, unsigned long event)
2410 {
2411         struct net_device_context *net_device_ctx;
2412         struct netvsc_device *netvsc_dev;
2413         struct net_device *ndev;
2414         bool vf_is_up = false;
2415         int ret;
2416
2417         if (event != NETDEV_GOING_DOWN)
2418                 vf_is_up = netif_running(vf_netdev);
2419
2420         ndev = get_netvsc_byref(vf_netdev);
2421         if (!ndev)
2422                 return NOTIFY_DONE;
2423
2424         net_device_ctx = netdev_priv(ndev);
2425         netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2426         if (!netvsc_dev)
2427                 return NOTIFY_DONE;
2428
2429         if (net_device_ctx->data_path_is_vf == vf_is_up)
2430                 return NOTIFY_OK;
2431
2432         ret = netvsc_switch_datapath(ndev, vf_is_up);
2433
2434         if (ret) {
2435                 netdev_err(ndev,
2436                            "Data path failed to switch %s VF: %s, err: %d\n",
2437                            vf_is_up ? "to" : "from", vf_netdev->name, ret);
2438                 return NOTIFY_DONE;
2439         } else {
2440                 netdev_info(ndev, "Data path switched %s VF: %s\n",
2441                             vf_is_up ? "to" : "from", vf_netdev->name);
2442         }
2443
2444         return NOTIFY_OK;
2445 }
2446
2447 static int netvsc_unregister_vf(struct net_device *vf_netdev)
2448 {
2449         struct net_device *ndev;
2450         struct net_device_context *net_device_ctx;
2451
2452         ndev = get_netvsc_byref(vf_netdev);
2453         if (!ndev)
2454                 return NOTIFY_DONE;
2455
2456         net_device_ctx = netdev_priv(ndev);
2457         cancel_delayed_work_sync(&net_device_ctx->vf_takeover);
2458
2459         netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
2460
2461         netvsc_vf_setxdp(vf_netdev, NULL);
2462
2463         netdev_rx_handler_unregister(vf_netdev);
2464         netdev_upper_dev_unlink(vf_netdev, ndev);
2465         RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
2466         dev_put(vf_netdev);
2467
2468         ndev->needed_headroom = RNDIS_AND_PPI_SIZE;
2469
2470         return NOTIFY_OK;
2471 }
2472
2473 static int netvsc_probe(struct hv_device *dev,
2474                         const struct hv_vmbus_device_id *dev_id)
2475 {
2476         struct net_device *net = NULL;
2477         struct net_device_context *net_device_ctx;
2478         struct netvsc_device_info *device_info = NULL;
2479         struct netvsc_device *nvdev;
2480         int ret = -ENOMEM;
2481
2482         net = alloc_etherdev_mq(sizeof(struct net_device_context),
2483                                 VRSS_CHANNEL_MAX);
2484         if (!net)
2485                 goto no_net;
2486
2487         netif_carrier_off(net);
2488
2489         netvsc_init_settings(net);
2490
2491         net_device_ctx = netdev_priv(net);
2492         net_device_ctx->device_ctx = dev;
2493         net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
2494         if (netif_msg_probe(net_device_ctx))
2495                 netdev_dbg(net, "netvsc msg_enable: %d\n",
2496                            net_device_ctx->msg_enable);
2497
2498         hv_set_drvdata(dev, net);
2499
2500         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
2501
2502         spin_lock_init(&net_device_ctx->lock);
2503         INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
2504         INIT_DELAYED_WORK(&net_device_ctx->vf_takeover, netvsc_vf_setup);
2505
2506         net_device_ctx->vf_stats
2507                 = netdev_alloc_pcpu_stats(struct netvsc_vf_pcpu_stats);
2508         if (!net_device_ctx->vf_stats)
2509                 goto no_stats;
2510
2511         net->netdev_ops = &device_ops;
2512         net->ethtool_ops = &ethtool_ops;
2513         SET_NETDEV_DEV(net, &dev->device);
2514
2515         /* We always need headroom for rndis header */
2516         net->needed_headroom = RNDIS_AND_PPI_SIZE;
2517
2518         /* Initialize the number of queues to be 1, we may change it if more
2519          * channels are offered later.
2520          */
2521         netif_set_real_num_tx_queues(net, 1);
2522         netif_set_real_num_rx_queues(net, 1);
2523
2524         /* Notify the netvsc driver of the new device */
2525         device_info = netvsc_devinfo_get(NULL);
2526
2527         if (!device_info) {
2528                 ret = -ENOMEM;
2529                 goto devinfo_failed;
2530         }
2531
2532         nvdev = rndis_filter_device_add(dev, device_info);
2533         if (IS_ERR(nvdev)) {
2534                 ret = PTR_ERR(nvdev);
2535                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
2536                 goto rndis_failed;
2537         }
2538
2539         memcpy(net->dev_addr, device_info->mac_adr, ETH_ALEN);
2540
2541         /* We must get rtnl lock before scheduling nvdev->subchan_work,
2542          * otherwise netvsc_subchan_work() can get rtnl lock first and wait
2543          * all subchannels to show up, but that may not happen because
2544          * netvsc_probe() can't get rtnl lock and as a result vmbus_onoffer()
2545          * -> ... -> device_add() -> ... -> __device_attach() can't get
2546          * the device lock, so all the subchannels can't be processed --
2547          * finally netvsc_subchan_work() hangs forever.
2548          */
2549         rtnl_lock();
2550
2551         if (nvdev->num_chn > 1)
2552                 schedule_work(&nvdev->subchan_work);
2553
2554         /* hw_features computed in rndis_netdev_set_hwcaps() */
2555         net->features = net->hw_features |
2556                 NETIF_F_HIGHDMA | NETIF_F_HW_VLAN_CTAG_TX |
2557                 NETIF_F_HW_VLAN_CTAG_RX;
2558         net->vlan_features = net->features;
2559
2560         netdev_lockdep_set_classes(net);
2561
2562         /* MTU range: 68 - 1500 or 65521 */
2563         net->min_mtu = NETVSC_MTU_MIN;
2564         if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
2565                 net->max_mtu = NETVSC_MTU - ETH_HLEN;
2566         else
2567                 net->max_mtu = ETH_DATA_LEN;
2568
2569         nvdev->tx_disable = false;
2570
2571         ret = register_netdevice(net);
2572         if (ret != 0) {
2573                 pr_err("Unable to register netdev.\n");
2574                 goto register_failed;
2575         }
2576
2577         list_add(&net_device_ctx->list, &netvsc_dev_list);
2578         rtnl_unlock();
2579
2580         netvsc_devinfo_put(device_info);
2581         return 0;
2582
2583 register_failed:
2584         rtnl_unlock();
2585         rndis_filter_device_remove(dev, nvdev);
2586 rndis_failed:
2587         netvsc_devinfo_put(device_info);
2588 devinfo_failed:
2589         free_percpu(net_device_ctx->vf_stats);
2590 no_stats:
2591         hv_set_drvdata(dev, NULL);
2592         free_netdev(net);
2593 no_net:
2594         return ret;
2595 }
2596
2597 static int netvsc_remove(struct hv_device *dev)
2598 {
2599         struct net_device_context *ndev_ctx;
2600         struct net_device *vf_netdev, *net;
2601         struct netvsc_device *nvdev;
2602
2603         net = hv_get_drvdata(dev);
2604         if (net == NULL) {
2605                 dev_err(&dev->device, "No net device to remove\n");
2606                 return 0;
2607         }
2608
2609         ndev_ctx = netdev_priv(net);
2610
2611         cancel_delayed_work_sync(&ndev_ctx->dwork);
2612
2613         rtnl_lock();
2614         nvdev = rtnl_dereference(ndev_ctx->nvdev);
2615         if (nvdev) {
2616                 cancel_work_sync(&nvdev->subchan_work);
2617                 netvsc_xdp_set(net, NULL, NULL, nvdev);
2618         }
2619
2620         /*
2621          * Call to the vsc driver to let it know that the device is being
2622          * removed. Also blocks mtu and channel changes.
2623          */
2624         vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2625         if (vf_netdev)
2626                 netvsc_unregister_vf(vf_netdev);
2627
2628         if (nvdev)
2629                 rndis_filter_device_remove(dev, nvdev);
2630
2631         unregister_netdevice(net);
2632         list_del(&ndev_ctx->list);
2633
2634         rtnl_unlock();
2635
2636         hv_set_drvdata(dev, NULL);
2637
2638         free_percpu(ndev_ctx->vf_stats);
2639         free_netdev(net);
2640         return 0;
2641 }
2642
2643 static int netvsc_suspend(struct hv_device *dev)
2644 {
2645         struct net_device_context *ndev_ctx;
2646         struct netvsc_device *nvdev;
2647         struct net_device *net;
2648         int ret;
2649
2650         net = hv_get_drvdata(dev);
2651
2652         ndev_ctx = netdev_priv(net);
2653         cancel_delayed_work_sync(&ndev_ctx->dwork);
2654
2655         rtnl_lock();
2656
2657         nvdev = rtnl_dereference(ndev_ctx->nvdev);
2658         if (nvdev == NULL) {
2659                 ret = -ENODEV;
2660                 goto out;
2661         }
2662
2663         /* Save the current config info */
2664         ndev_ctx->saved_netvsc_dev_info = netvsc_devinfo_get(nvdev);
2665
2666         ret = netvsc_detach(net, nvdev);
2667 out:
2668         rtnl_unlock();
2669
2670         return ret;
2671 }
2672
2673 static int netvsc_resume(struct hv_device *dev)
2674 {
2675         struct net_device *net = hv_get_drvdata(dev);
2676         struct net_device_context *net_device_ctx;
2677         struct netvsc_device_info *device_info;
2678         int ret;
2679
2680         rtnl_lock();
2681
2682         net_device_ctx = netdev_priv(net);
2683
2684         /* Reset the data path to the netvsc NIC before re-opening the vmbus
2685          * channel. Later netvsc_netdev_event() will switch the data path to
2686          * the VF upon the UP or CHANGE event.
2687          */
2688         net_device_ctx->data_path_is_vf = false;
2689         device_info = net_device_ctx->saved_netvsc_dev_info;
2690
2691         ret = netvsc_attach(net, device_info);
2692
2693         netvsc_devinfo_put(device_info);
2694         net_device_ctx->saved_netvsc_dev_info = NULL;
2695
2696         rtnl_unlock();
2697
2698         return ret;
2699 }
2700 static const struct hv_vmbus_device_id id_table[] = {
2701         /* Network guid */
2702         { HV_NIC_GUID, },
2703         { },
2704 };
2705
2706 MODULE_DEVICE_TABLE(vmbus, id_table);
2707
2708 /* The one and only one */
2709 static struct  hv_driver netvsc_drv = {
2710         .name = KBUILD_MODNAME,
2711         .id_table = id_table,
2712         .probe = netvsc_probe,
2713         .remove = netvsc_remove,
2714         .suspend = netvsc_suspend,
2715         .resume = netvsc_resume,
2716         .driver = {
2717                 .probe_type = PROBE_FORCE_SYNCHRONOUS,
2718         },
2719 };
2720
2721 /*
2722  * On Hyper-V, every VF interface is matched with a corresponding
2723  * synthetic interface. The synthetic interface is presented first
2724  * to the guest. When the corresponding VF instance is registered,
2725  * we will take care of switching the data path.
2726  */
2727 static int netvsc_netdev_event(struct notifier_block *this,
2728                                unsigned long event, void *ptr)
2729 {
2730         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2731
2732         /* Skip our own events */
2733         if (event_dev->netdev_ops == &device_ops)
2734                 return NOTIFY_DONE;
2735
2736         /* Avoid non-Ethernet type devices */
2737         if (event_dev->type != ARPHRD_ETHER)
2738                 return NOTIFY_DONE;
2739
2740         /* Avoid Vlan dev with same MAC registering as VF */
2741         if (is_vlan_dev(event_dev))
2742                 return NOTIFY_DONE;
2743
2744         /* Avoid Bonding master dev with same MAC registering as VF */
2745         if ((event_dev->priv_flags & IFF_BONDING) &&
2746             (event_dev->flags & IFF_MASTER))
2747                 return NOTIFY_DONE;
2748
2749         switch (event) {
2750         case NETDEV_REGISTER:
2751                 return netvsc_register_vf(event_dev);
2752         case NETDEV_UNREGISTER:
2753                 return netvsc_unregister_vf(event_dev);
2754         case NETDEV_UP:
2755         case NETDEV_DOWN:
2756         case NETDEV_CHANGE:
2757         case NETDEV_GOING_DOWN:
2758                 return netvsc_vf_changed(event_dev, event);
2759         default:
2760                 return NOTIFY_DONE;
2761         }
2762 }
2763
2764 static struct notifier_block netvsc_netdev_notifier = {
2765         .notifier_call = netvsc_netdev_event,
2766 };
2767
2768 static void __exit netvsc_drv_exit(void)
2769 {
2770         unregister_netdevice_notifier(&netvsc_netdev_notifier);
2771         vmbus_driver_unregister(&netvsc_drv);
2772 }
2773
2774 static int __init netvsc_drv_init(void)
2775 {
2776         int ret;
2777
2778         if (ring_size < RING_SIZE_MIN) {
2779                 ring_size = RING_SIZE_MIN;
2780                 pr_info("Increased ring_size to %u (min allowed)\n",
2781                         ring_size);
2782         }
2783         netvsc_ring_bytes = ring_size * PAGE_SIZE;
2784
2785         ret = vmbus_driver_register(&netvsc_drv);
2786         if (ret)
2787                 return ret;
2788
2789         register_netdevice_notifier(&netvsc_netdev_notifier);
2790         return 0;
2791 }
2792
2793 MODULE_LICENSE("GPL");
2794 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
2795
2796 module_init(netvsc_drv_init);
2797 module_exit(netvsc_drv_exit);