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