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 <[email protected]>
   7  *   Hank Janssen  <[email protected]>
   8  */
   9 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  10
  11 #include <linux/init.h>
  12 #include <linux/atomic.h>
  13 #include <linux/module.h>
  14 #include <linux/highmem.h>
  15 #include <linux/device.h>
  16 #include <linux/io.h>
  17 #include <linux/delay.h>
  18 #include <linux/netdevice.h>
  19 #include <linux/inetdevice.h>
  20 #include <linux/etherdevice.h>
  21 #include <linux/pci.h>
  22 #include <linux/skbuff.h>
  23 #include <linux/if_vlan.h>
  24 #include <linux/in.h>
  25 #include <linux/slab.h>
  26 #include <linux/rtnetlink.h>
  27 #include <linux/netpoll.h>
  28 #include <linux/bpf.h>
  29
  30 #include <net/arp.h>
  31 #include <net/route.h>
  32 #include <net/sock.h>
  33 #include <net/pkt_sched.h>
  34 #include <net/checksum.h>
  35 #include <net/ip6_checksum.h>
  36
  37 #include "hyperv_net.h"
  38
  39 #define RING_SIZE_MIN   64
  40 #define RETRY_US_LO     5000
  41 #define RETRY_US_HI     10000
  42 #define RETRY_MAX       2000    /* >10 sec */
  43
  44 #define LINKCHANGE_INT (2 * HZ)
  45 #define VF_TAKEOVER_INT (HZ / 10)
  46
  47 static unsigned int ring_size __ro_after_init = 128;
  48 module_param(ring_size, uint, 0444);
  49 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
  50 unsigned int netvsc_ring_bytes __ro_after_init;
  51
  52 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
  53                                 NETIF_MSG_LINK | NETIF_MSG_IFUP |
  54                                 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
  55                                 NETIF_MSG_TX_ERR;
  56
  57 static int debug = -1;
  58 module_param(debug, int, 0444);
  59 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  60
  61 static LIST_HEAD(netvsc_dev_list);
  62
  63 static void netvsc_change_rx_flags(struct net_device *net, int change)
  64 {
  65         struct net_device_context *ndev_ctx = netdev_priv(net);
  66         struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
  67         int inc;
  68
  69         if (!vf_netdev)
  70                 return;
  71
  72         if (change & IFF_PROMISC) {
  73                 inc = (net->flags & IFF_PROMISC) ? 1 : -1;
  74                 dev_set_promiscuity(vf_netdev, inc);
  75         }
  76
  77         if (change & IFF_ALLMULTI) {
  78                 inc = (net->flags & IFF_ALLMULTI) ? 1 : -1;
  79                 dev_set_allmulti(vf_netdev, inc);
  80         }
  81 }
  82
  83 static void netvsc_set_rx_mode(struct net_device *net)
  84 {
  85         struct net_device_context *ndev_ctx = netdev_priv(net);
  86         struct net_device *vf_netdev;
  87         struct netvsc_device *nvdev;
  88
  89         rcu_read_lock();
  90         vf_netdev = rcu_dereference(ndev_ctx->vf_netdev);
  91         if (vf_netdev) {
  92                 dev_uc_sync(vf_netdev, net);
  93                 dev_mc_sync(vf_netdev, net);
  94         }
  95
  96         nvdev = rcu_dereference(ndev_ctx->nvdev);
  97         if (nvdev)
  98                 rndis_filter_update(nvdev);
  99         rcu_read_unlock();
 100 }
 101
 102 static void netvsc_tx_enable(struct netvsc_device *nvscdev,
 103                              struct net_device *ndev)
 104 {
 105         nvscdev->tx_disable = false;
 106         virt_wmb(); /* ensure queue wake up mechanism is on */
 107
 108         netif_tx_wake_all_queues(ndev);
 109 }
 110
 111 static int netvsc_open(struct net_device *net)
 112 {
 113         struct net_device_context *ndev_ctx = netdev_priv(net);
 114         struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
 115         struct netvsc_device *nvdev = rtnl_dereference(ndev_ctx->nvdev);
 116         struct rndis_device *rdev;
 117         int ret = 0;
 118
 119         netif_carrier_off(net);
 120
 121         /* Open up the device */
 122         ret = rndis_filter_open(nvdev);
 123         if (ret != 0) {
 124                 netdev_err(net, "unable to open device (ret %d).\n", ret);
 125                 return ret;
 126         }
 127
 128         rdev = nvdev->extension;
 129         if (!rdev->link_state) {
 130                 netif_carrier_on(net);
 131                 netvsc_tx_enable(nvdev, net);
 132         }
 133
 134         if (vf_netdev) {
 135                 /* Setting synthetic device up transparently sets
 136                  * slave as up. If open fails, then slave will be
 137                  * still be offline (and not used).
 138                  */
 139                 ret = dev_open(vf_netdev, NULL);
 140                 if (ret)
 141                         netdev_warn(net,
 142                                     "unable to open slave: %s: %d\n",
 143                                     vf_netdev->name, ret);
 144         }
 145         return 0;
 146 }
 147
 148 static int netvsc_wait_until_empty(struct netvsc_device *nvdev)
 149 {
 150         unsigned int retry = 0;
 151         int i;
 152
 153         /* Ensure pending bytes in ring are read */
 154         for (;;) {
 155                 u32 aread = 0;
 156
 157                 for (i = 0; i < nvdev->num_chn; i++) {
 158                         struct vmbus_channel *chn
 159                                 = nvdev->chan_table[i].channel;
 160
 161                         if (!chn)
 162                                 continue;
 163
 164                         /* make sure receive not running now */
 165                         napi_synchronize(&nvdev->chan_table[i].napi);
 166
 167                         aread = hv_get_bytes_to_read(&chn->inbound);
 168                         if (aread)
 169                                 break;
 170
 171                         aread = hv_get_bytes_to_read(&chn->outbound);
 172                         if (aread)
 173                                 break;
 174                 }
 175
 176                 if (aread == 0)
 177                         return 0;
 178
 179                 if (++retry > RETRY_MAX)
 180                         return -ETIMEDOUT;
 181
 182                 usleep_range(RETRY_US_LO, RETRY_US_HI);
 183         }
 184 }
 185
 186 static void netvsc_tx_disable(struct netvsc_device *nvscdev,
 187                               struct net_device *ndev)
 188 {
 189         if (nvscdev) {
 190                 nvscdev->tx_disable = true;
 191                 virt_wmb(); /* ensure txq will not wake up after stop */
 192         }
 193
 194         netif_tx_disable(ndev);
 195 }
 196
 197 static int netvsc_close(struct net_device *net)
 198 {
 199         struct net_device_context *net_device_ctx = netdev_priv(net);
 200         struct net_device *vf_netdev
 201                 = rtnl_dereference(net_device_ctx->vf_netdev);
 202         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
 203         int ret;
 204
 205         netvsc_tx_disable(nvdev, net);
 206
 207         /* No need to close rndis filter if it is removed already */
 208         if (!nvdev)
 209                 return 0;
 210
 211         ret = rndis_filter_close(nvdev);
 212         if (ret != 0) {
 213                 netdev_err(net, "unable to close device (ret %d).\n", ret);
 214                 return ret;
 215         }
 216
 217         ret = netvsc_wait_until_empty(nvdev);
 218         if (ret)
 219                 netdev_err(net, "Ring buffer not empty after closing rndis\n");
 220
 221         if (vf_netdev)
 222                 dev_close(vf_netdev);
 223
 224         return ret;
 225 }
 226
 227 static inline void *init_ppi_data(struct rndis_message *msg,
 228                                   u32 ppi_size, u32 pkt_type)
 229 {
 230         struct rndis_packet *rndis_pkt = &msg->msg.pkt;
 231         struct rndis_per_packet_info *ppi;
 232
 233         rndis_pkt->data_offset += ppi_size;
 234         ppi = (void *)rndis_pkt + rndis_pkt->per_pkt_info_offset
 235                 + rndis_pkt->per_pkt_info_len;
 236
 237         ppi->size = ppi_size;
 238         ppi->type = pkt_type;
 239         ppi->internal = 0;
 240         ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
 241
 242         rndis_pkt->per_pkt_info_len += ppi_size;
 243
 244         return ppi + 1;
 245 }
 246
 247 /* Azure hosts don't support non-TCP port numbers in hashing for fragmented
 248  * packets. We can use ethtool to change UDP hash level when necessary.
 249  */
 250 static inline u32 netvsc_get_hash(
 251         struct sk_buff *skb,
 252         const struct net_device_context *ndc)
 253 {
 254         struct flow_keys flow;
 255         u32 hash, pkt_proto = 0;
 256         static u32 hashrnd __read_mostly;
 257
 258         net_get_random_once(&hashrnd, sizeof(hashrnd));
 259
 260         if (!skb_flow_dissect_flow_keys(skb, &flow, 0))
 261                 return 0;
 262
 263         switch (flow.basic.ip_proto) {
 264         case IPPROTO_TCP:
 265                 if (flow.basic.n_proto == htons(ETH_P_IP))
 266                         pkt_proto = HV_TCP4_L4HASH;
 267                 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
 268                         pkt_proto = HV_TCP6_L4HASH;
 269
 270                 break;
 271
 272         case IPPROTO_UDP:
 273                 if (flow.basic.n_proto == htons(ETH_P_IP))
 274                         pkt_proto = HV_UDP4_L4HASH;
 275                 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
 276                         pkt_proto = HV_UDP6_L4HASH;
 277
 278                 break;
 279         }
 280
 281         if (pkt_proto & ndc->l4_hash) {
 282                 return skb_get_hash(skb);
 283         } else {
 284                 if (flow.basic.n_proto == htons(ETH_P_IP))
 285                         hash = jhash2((u32 *)&flow.addrs.v4addrs, 2, hashrnd);
 286                 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
 287                         hash = jhash2((u32 *)&flow.addrs.v6addrs, 8, hashrnd);
 288                 else
 289                         return 0;
 290
 291                 __skb_set_sw_hash(skb, hash, false);
 292         }
 293
 294         return hash;
 295 }
 296
 297 static inline int netvsc_get_tx_queue(struct net_device *ndev,
 298                                       struct sk_buff *skb, int old_idx)
 299 {
 300         const struct net_device_context *ndc = netdev_priv(ndev);
 301         struct sock *sk = skb->sk;
 302         int q_idx;
 303
 304         q_idx = ndc->tx_table[netvsc_get_hash(skb, ndc) &
 305                               (VRSS_SEND_TAB_SIZE - 1)];
 306
 307         /* If queue index changed record the new value */
 308         if (q_idx != old_idx &&
 309             sk && sk_fullsock(sk) && rcu_access_pointer(sk->sk_dst_cache))
 310                 sk_tx_queue_set(sk, q_idx);
 311
 312         return q_idx;
 313 }
 314
 315 /*
 316  * Select queue for transmit.
 317  *
 318  * If a valid queue has already been assigned, then use that.
 319  * Otherwise compute tx queue based on hash and the send table.
 320  *
 321  * This is basically similar to default (netdev_pick_tx) with the added step
 322  * of using the host send_table when no other queue has been assigned.
 323  *
 324  * TODO support XPS - but get_xps_queue not exported
 325  */
 326 static u16 netvsc_pick_tx(struct net_device *ndev, struct sk_buff *skb)
 327 {
 328         int q_idx = sk_tx_queue_get(skb->sk);
 329
 330         if (q_idx < 0 || skb->ooo_okay || q_idx >= ndev->real_num_tx_queues) {
 331                 /* If forwarding a packet, we use the recorded queue when
 332                  * available for better cache locality.
 333                  */
 334                 if (skb_rx_queue_recorded(skb))
 335                         q_idx = skb_get_rx_queue(skb);
 336                 else
 337                         q_idx = netvsc_get_tx_queue(ndev, skb, q_idx);
 338         }
 339
 340         return q_idx;
 341 }
 342
 343 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
 344                                struct net_device *sb_dev)
 345 {
 346         struct net_device_context *ndc = netdev_priv(ndev);
 347         struct net_device *vf_netdev;
 348         u16 txq;
 349
 350         rcu_read_lock();
 351         vf_netdev = rcu_dereference(ndc->vf_netdev);
 352         if (vf_netdev) {
 353                 const struct net_device_ops *vf_ops = vf_netdev->netdev_ops;
 354
 355                 if (vf_ops->ndo_select_queue)
 356                         txq = vf_ops->ndo_select_queue(vf_netdev, skb, sb_dev);
 357                 else
 358                         txq = netdev_pick_tx(vf_netdev, skb, NULL);
 359
 360                 /* Record the queue selected by VF so that it can be
 361                  * used for common case where VF has more queues than
 362                  * the synthetic device.
 363                  */
 364                 qdisc_skb_cb(skb)->slave_dev_queue_mapping = txq;
 365         } else {
 366                 txq = netvsc_pick_tx(ndev, skb);
 367         }
 368         rcu_read_unlock();
 369
 370         while (unlikely(txq >= ndev->real_num_tx_queues))
 371                 txq -= ndev->real_num_tx_queues;
 372
 373         return txq;
 374 }
 375
 376 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
 377                        struct hv_page_buffer *pb)
 378 {
 379         int j = 0;
 380
 381         /* Deal with compound pages by ignoring unused part
 382          * of the page.
 383          */
 384         page += (offset >> PAGE_SHIFT);
 385         offset &= ~PAGE_MASK;
 386
 387         while (len > 0) {
 388                 unsigned long bytes;
 389
 390                 bytes = PAGE_SIZE - offset;
 391                 if (bytes > len)
 392                         bytes = len;
 393                 pb[j].pfn = page_to_pfn(page);
 394                 pb[j].offset = offset;
 395                 pb[j].len = bytes;
 396
 397                 offset += bytes;
 398                 len -= bytes;
 399
 400                 if (offset == PAGE_SIZE && len) {
 401                         page++;
 402                         offset = 0;
 403                         j++;
 404                 }
 405         }
 406
 407         return j + 1;
 408 }
 409
 410 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
 411                            struct hv_netvsc_packet *packet,
 412                            struct hv_page_buffer *pb)
 413 {
 414         u32 slots_used = 0;
 415         char *data = skb->data;
 416         int frags = skb_shinfo(skb)->nr_frags;
 417         int i;
 418
 419         /* The packet is laid out thus:
 420          * 1. hdr: RNDIS header and PPI
 421          * 2. skb linear data
 422          * 3. skb fragment data
 423          */
 424         slots_used += fill_pg_buf(virt_to_page(hdr),
 425                                   offset_in_page(hdr),
 426                                   len, &pb[slots_used]);
 427
 428         packet->rmsg_size = len;
 429         packet->rmsg_pgcnt = slots_used;
 430
 431         slots_used += fill_pg_buf(virt_to_page(data),
 432                                 offset_in_page(data),
 433                                 skb_headlen(skb), &pb[slots_used]);
 434
 435         for (i = 0; i < frags; i++) {
 436                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 437
 438                 slots_used += fill_pg_buf(skb_frag_page(frag),
 439                                         skb_frag_off(frag),
 440                                         skb_frag_size(frag), &pb[slots_used]);
 441         }
 442         return slots_used;
 443 }
 444
 445 static int count_skb_frag_slots(struct sk_buff *skb)
 446 {
 447         int i, frags = skb_shinfo(skb)->nr_frags;
 448         int pages = 0;
 449
 450         for (i = 0; i < frags; i++) {
 451                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 452                 unsigned long size = skb_frag_size(frag);
 453                 unsigned long offset = skb_frag_off(frag);
 454
 455                 /* Skip unused frames from start of page */
 456                 offset &= ~PAGE_MASK;
 457                 pages += PFN_UP(offset + size);
 458         }
 459         return pages;
 460 }
 461
 462 static int netvsc_get_slots(struct sk_buff *skb)
 463 {
 464         char *data = skb->data;
 465         unsigned int offset = offset_in_page(data);
 466         unsigned int len = skb_headlen(skb);
 467         int slots;
 468         int frag_slots;
 469
 470         slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
 471         frag_slots = count_skb_frag_slots(skb);
 472         return slots + frag_slots;
 473 }
 474
 475 static u32 net_checksum_info(struct sk_buff *skb)
 476 {
 477         if (skb->protocol == htons(ETH_P_IP)) {
 478                 struct iphdr *ip = ip_hdr(skb);
 479
 480                 if (ip->protocol == IPPROTO_TCP)
 481                         return TRANSPORT_INFO_IPV4_TCP;
 482                 else if (ip->protocol == IPPROTO_UDP)
 483                         return TRANSPORT_INFO_IPV4_UDP;
 484         } else {
 485                 struct ipv6hdr *ip6 = ipv6_hdr(skb);
 486
 487                 if (ip6->nexthdr == IPPROTO_TCP)
 488                         return TRANSPORT_INFO_IPV6_TCP;
 489                 else if (ip6->nexthdr == IPPROTO_UDP)
 490                         return TRANSPORT_INFO_IPV6_UDP;
 491         }
 492
 493         return TRANSPORT_INFO_NOT_IP;
 494 }
 495
 496 /* Send skb on the slave VF device. */
 497 static int netvsc_vf_xmit(struct net_device *net, struct net_device *vf_netdev,
 498                           struct sk_buff *skb)
 499 {
 500         struct net_device_context *ndev_ctx = netdev_priv(net);
 501         unsigned int len = skb->len;
 502         int rc;
 503
 504         skb->dev = vf_netdev;
 505         skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
 506
 507         rc = dev_queue_xmit(skb);
 508         if (likely(rc == NET_XMIT_SUCCESS || rc == NET_XMIT_CN)) {
 509                 struct netvsc_vf_pcpu_stats *pcpu_stats
 510                         = this_cpu_ptr(ndev_ctx->vf_stats);
 511
 512                 u64_stats_update_begin(&pcpu_stats->syncp);
 513                 pcpu_stats->tx_packets++;
 514                 pcpu_stats->tx_bytes += len;
 515                 u64_stats_update_end(&pcpu_stats->syncp);
 516         } else {
 517                 this_cpu_inc(ndev_ctx->vf_stats->tx_dropped);
 518         }
 519
 520         return rc;
 521 }
 522
 523 static int netvsc_xmit(struct sk_buff *skb, struct net_device *net, bool xdp_tx)
 524 {
 525         struct net_device_context *net_device_ctx = netdev_priv(net);
 526         struct hv_netvsc_packet *packet = NULL;
 527         int ret;
 528         unsigned int num_data_pgs;
 529         struct rndis_message *rndis_msg;
 530         struct net_device *vf_netdev;
 531         u32 rndis_msg_size;
 532         u32 hash;
 533         struct hv_page_buffer pb[MAX_PAGE_BUFFER_COUNT];
 534
 535         /* If VF is present and up then redirect packets to it.
 536          * Skip the VF if it is marked down or has no carrier.
 537          * If netpoll is in uses, then VF can not be used either.
 538          */
 539         vf_netdev = rcu_dereference_bh(net_device_ctx->vf_netdev);
 540         if (vf_netdev && netif_running(vf_netdev) &&
 541             netif_carrier_ok(vf_netdev) && !netpoll_tx_running(net))
 542                 return netvsc_vf_xmit(net, vf_netdev, skb);
 543
 544         /* We will atmost need two pages to describe the rndis
 545          * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
 546          * of pages in a single packet. If skb is scattered around
 547          * more pages we try linearizing it.
 548          */
 549
 550         num_data_pgs = netvsc_get_slots(skb) + 2;
 551
 552         if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
 553                 ++net_device_ctx->eth_stats.tx_scattered;
 554
 555                 if (skb_linearize(skb))
 556                         goto no_memory;
 557
 558                 num_data_pgs = netvsc_get_slots(skb) + 2;
 559                 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
 560                         ++net_device_ctx->eth_stats.tx_too_big;
 561                         goto drop;
 562                 }
 563         }
 564
 565         /*
 566          * Place the rndis header in the skb head room and
 567          * the skb->cb will be used for hv_netvsc_packet
 568          * structure.
 569          */
 570         ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
 571         if (ret)
 572                 goto no_memory;
 573
 574         /* Use the skb control buffer for building up the packet */
 575         BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
 576                         sizeof_field(struct sk_buff, cb));
 577         packet = (struct hv_netvsc_packet *)skb->cb;
 578
 579         packet->q_idx = skb_get_queue_mapping(skb);
 580
 581         packet->total_data_buflen = skb->len;
 582         packet->total_bytes = skb->len;
 583         packet->total_packets = 1;
 584
 585         rndis_msg = (struct rndis_message *)skb->head;
 586
 587         /* Add the rndis header */
 588         rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
 589         rndis_msg->msg_len = packet->total_data_buflen;
 590
 591         rndis_msg->msg.pkt = (struct rndis_packet) {
 592                 .data_offset = sizeof(struct rndis_packet),
 593                 .data_len = packet->total_data_buflen,
 594                 .per_pkt_info_offset = sizeof(struct rndis_packet),
 595         };
 596
 597         rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
 598
 599         hash = skb_get_hash_raw(skb);
 600         if (hash != 0 && net->real_num_tx_queues > 1) {
 601                 u32 *hash_info;
 602
 603                 rndis_msg_size += NDIS_HASH_PPI_SIZE;
 604                 hash_info = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
 605                                           NBL_HASH_VALUE);
 606                 *hash_info = hash;
 607         }
 608
 609         /* When using AF_PACKET we need to drop VLAN header from
 610          * the frame and update the SKB to allow the HOST OS
 611          * to transmit the 802.1Q packet
 612          */
 613         if (skb->protocol == htons(ETH_P_8021Q)) {
 614                 u16 vlan_tci;
 615
 616                 skb_reset_mac_header(skb);
 617                 if (eth_type_vlan(eth_hdr(skb)->h_proto)) {
 618                         if (unlikely(__skb_vlan_pop(skb, &vlan_tci) != 0)) {
 619                                 ++net_device_ctx->eth_stats.vlan_error;
 620                                 goto drop;
 621                         }
 622
 623                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci);
 624                         /* Update the NDIS header pkt lengths */
 625                         packet->total_data_buflen -= VLAN_HLEN;
 626                         packet->total_bytes -= VLAN_HLEN;
 627                         rndis_msg->msg_len = packet->total_data_buflen;
 628                         rndis_msg->msg.pkt.data_len = packet->total_data_buflen;
 629                 }
 630         }
 631
 632         if (skb_vlan_tag_present(skb)) {
 633                 struct ndis_pkt_8021q_info *vlan;
 634
 635                 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
 636                 vlan = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
 637                                      IEEE_8021Q_INFO);
 638
 639                 vlan->value = 0;
 640                 vlan->vlanid = skb_vlan_tag_get_id(skb);
 641                 vlan->cfi = skb_vlan_tag_get_cfi(skb);
 642                 vlan->pri = skb_vlan_tag_get_prio(skb);
 643         }
 644
 645         if (skb_is_gso(skb)) {
 646                 struct ndis_tcp_lso_info *lso_info;
 647
 648                 rndis_msg_size += NDIS_LSO_PPI_SIZE;
 649                 lso_info = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
 650                                          TCP_LARGESEND_PKTINFO);
 651
 652                 lso_info->value = 0;
 653                 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
 654                 if (skb->protocol == htons(ETH_P_IP)) {
 655                         lso_info->lso_v2_transmit.ip_version =
 656                                 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
 657                         ip_hdr(skb)->tot_len = 0;
 658                         ip_hdr(skb)->check = 0;
 659                         tcp_hdr(skb)->check =
 660                                 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
 661                                                    ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 662                 } else {
 663                         lso_info->lso_v2_transmit.ip_version =
 664                                 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
 665                         tcp_v6_gso_csum_prep(skb);
 666                 }
 667                 lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb);
 668                 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
 669         } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
 670                 if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) {
 671                         struct ndis_tcp_ip_checksum_info *csum_info;
 672
 673                         rndis_msg_size += NDIS_CSUM_PPI_SIZE;
 674                         csum_info = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
 675                                                   TCPIP_CHKSUM_PKTINFO);
 676
 677                         csum_info->value = 0;
 678                         csum_info->transmit.tcp_header_offset = skb_transport_offset(skb);
 679
 680                         if (skb->protocol == htons(ETH_P_IP)) {
 681                                 csum_info->transmit.is_ipv4 = 1;
 682
 683                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
 684                                         csum_info->transmit.tcp_checksum = 1;
 685                                 else
 686                                         csum_info->transmit.udp_checksum = 1;
 687                         } else {
 688                                 csum_info->transmit.is_ipv6 = 1;
 689
 690                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
 691                                         csum_info->transmit.tcp_checksum = 1;
 692                                 else
 693                                         csum_info->transmit.udp_checksum = 1;
 694                         }
 695                 } else {
 696                         /* Can't do offload of this type of checksum */
 697                         if (skb_checksum_help(skb))
 698                                 goto drop;
 699                 }
 700         }
 701
 702         /* Start filling in the page buffers with the rndis hdr */
 703         rndis_msg->msg_len += rndis_msg_size;
 704         packet->total_data_buflen = rndis_msg->msg_len;
 705         packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
 706                                                skb, packet, pb);
 707
 708         /* timestamp packet in software */
 709         skb_tx_timestamp(skb);
 710
 711         ret = netvsc_send(net, packet, rndis_msg, pb, skb, xdp_tx);
 712         if (likely(ret == 0))
 713                 return NETDEV_TX_OK;
 714
 715         if (ret == -EAGAIN) {
 716                 ++net_device_ctx->eth_stats.tx_busy;
 717                 return NETDEV_TX_BUSY;
 718         }
 719
 720         if (ret == -ENOSPC)
 721                 ++net_device_ctx->eth_stats.tx_no_space;
 722
 723 drop:
 724         dev_kfree_skb_any(skb);
 725         net->stats.tx_dropped++;
 726
 727         return NETDEV_TX_OK;
 728
 729 no_memory:
 730         ++net_device_ctx->eth_stats.tx_no_memory;
 731         goto drop;
 732 }
 733
 734 static netdev_tx_t netvsc_start_xmit(struct sk_buff *skb,
 735                                      struct net_device *ndev)
 736 {
 737         return netvsc_xmit(skb, ndev, false);
 738 }
 739
 740 /*
 741  * netvsc_linkstatus_callback - Link up/down notification
 742  */
 743 void netvsc_linkstatus_callback(struct net_device *net,
 744                                 struct rndis_message *resp)
 745 {
 746         struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
 747         struct net_device_context *ndev_ctx = netdev_priv(net);
 748         struct netvsc_reconfig *event;
 749         unsigned long flags;
 750
 751         /* Update the physical link speed when changing to another vSwitch */
 752         if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
 753                 u32 speed;
 754
 755                 speed = *(u32 *)((void *)indicate
 756                                  + indicate->status_buf_offset) / 10000;
 757                 ndev_ctx->speed = speed;
 758                 return;
 759         }
 760
 761         /* Handle these link change statuses below */
 762         if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
 763             indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
 764             indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
 765                 return;
 766
 767         if (net->reg_state != NETREG_REGISTERED)
 768                 return;
 769
 770         event = kzalloc(sizeof(*event), GFP_ATOMIC);
 771         if (!event)
 772                 return;
 773         event->event = indicate->status;
 774
 775         spin_lock_irqsave(&ndev_ctx->lock, flags);
 776         list_add_tail(&event->list, &ndev_ctx->reconfig_events);
 777         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
 778
 779         schedule_delayed_work(&ndev_ctx->dwork, 0);
 780 }
 781
 782 static void netvsc_xdp_xmit(struct sk_buff *skb, struct net_device *ndev)
 783 {
 784         int rc;
 785
 786         skb->queue_mapping = skb_get_rx_queue(skb);
 787         __skb_push(skb, ETH_HLEN);
 788
 789         rc = netvsc_xmit(skb, ndev, true);
 790
 791         if (dev_xmit_complete(rc))
 792                 return;
 793
 794         dev_kfree_skb_any(skb);
 795         ndev->stats.tx_dropped++;
 796 }
 797
 798 static void netvsc_comp_ipcsum(struct sk_buff *skb)
 799 {
 800         struct iphdr *iph = (struct iphdr *)skb->data;
 801
 802         iph->check = 0;
 803         iph->check = ip_fast_csum(iph, iph->ihl);
 804 }
 805
 806 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
 807                                              struct netvsc_channel *nvchan,
 808                                              struct xdp_buff *xdp)
 809 {
 810         struct napi_struct *napi = &nvchan->napi;
 811         const struct ndis_pkt_8021q_info *vlan = nvchan->rsc.vlan;
 812         const struct ndis_tcp_ip_checksum_info *csum_info =
 813                                                 nvchan->rsc.csum_info;
 814         const u32 *hash_info = nvchan->rsc.hash_info;
 815         struct sk_buff *skb;
 816         void *xbuf = xdp->data_hard_start;
 817         int i;
 818
 819         if (xbuf) {
 820                 unsigned int hdroom = xdp->data - xdp->data_hard_start;
 821                 unsigned int xlen = xdp->data_end - xdp->data;
 822                 unsigned int frag_size = xdp->frame_sz;
 823
 824                 skb = build_skb(xbuf, frag_size);
 825
 826                 if (!skb) {
 827                         __free_page(virt_to_page(xbuf));
 828                         return NULL;
 829                 }
 830
 831                 skb_reserve(skb, hdroom);
 832                 skb_put(skb, xlen);
 833                 skb->dev = napi->dev;
 834         } else {
 835                 skb = napi_alloc_skb(napi, nvchan->rsc.pktlen);
 836
 837                 if (!skb)
 838                         return NULL;
 839
 840                 /* Copy to skb. This copy is needed here since the memory
 841                  * pointed by hv_netvsc_packet cannot be deallocated.
 842                  */
 843                 for (i = 0; i < nvchan->rsc.cnt; i++)
 844                         skb_put_data(skb, nvchan->rsc.data[i],
 845                                      nvchan->rsc.len[i]);
 846         }
 847
 848         skb->protocol = eth_type_trans(skb, net);
 849
 850         /* skb is already created with CHECKSUM_NONE */
 851         skb_checksum_none_assert(skb);
 852
 853         /* Incoming packets may have IP header checksum verified by the host.
 854          * They may not have IP header checksum computed after coalescing.
 855          * We compute it here if the flags are set, because on Linux, the IP
 856          * checksum is always checked.
 857          */
 858         if (csum_info && csum_info->receive.ip_checksum_value_invalid &&
 859             csum_info->receive.ip_checksum_succeeded &&
 860             skb->protocol == htons(ETH_P_IP))
 861                 netvsc_comp_ipcsum(skb);
 862
 863         /* Do L4 checksum offload if enabled and present. */
 864         if (csum_info && (net->features & NETIF_F_RXCSUM)) {
 865                 if (csum_info->receive.tcp_checksum_succeeded ||
 866                     csum_info->receive.udp_checksum_succeeded)
 867                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 868         }
 869
 870         if (hash_info && (net->features & NETIF_F_RXHASH))
 871                 skb_set_hash(skb, *hash_info, PKT_HASH_TYPE_L4);
 872
 873         if (vlan) {
 874                 u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT) |
 875                         (vlan->cfi ? VLAN_CFI_MASK : 0);
 876
 877                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 878                                        vlan_tci);
 879         }
 880
 881         return skb;
 882 }
 883
 884 /*
 885  * netvsc_recv_callback -  Callback when we receive a packet from the
 886  * "wire" on the specified device.
 887  */
 888 int netvsc_recv_callback(struct net_device *net,
 889                          struct netvsc_device *net_device,
 890                          struct netvsc_channel *nvchan)
 891 {
 892         struct net_device_context *net_device_ctx = netdev_priv(net);
 893         struct vmbus_channel *channel = nvchan->channel;
 894         u16 q_idx = channel->offermsg.offer.sub_channel_index;
 895         struct sk_buff *skb;
 896         struct netvsc_stats *rx_stats = &nvchan->rx_stats;
 897         struct xdp_buff xdp;
 898         u32 act;
 899
 900         if (net->reg_state != NETREG_REGISTERED)
 901                 return NVSP_STAT_FAIL;
 902
 903         act = netvsc_run_xdp(net, nvchan, &xdp);
 904
 905         if (act != XDP_PASS && act != XDP_TX) {
 906                 u64_stats_update_begin(&rx_stats->syncp);
 907                 rx_stats->xdp_drop++;
 908                 u64_stats_update_end(&rx_stats->syncp);
 909
 910                 return NVSP_STAT_SUCCESS; /* consumed by XDP */
 911         }
 912
 913         /* Allocate a skb - TODO direct I/O to pages? */
 914         skb = netvsc_alloc_recv_skb(net, nvchan, &xdp);
 915
 916         if (unlikely(!skb)) {
 917                 ++net_device_ctx->eth_stats.rx_no_memory;
 918                 return NVSP_STAT_FAIL;
 919         }
 920
 921         skb_record_rx_queue(skb, q_idx);
 922
 923         /*
 924          * Even if injecting the packet, record the statistics
 925          * on the synthetic device because modifying the VF device
 926          * statistics will not work correctly.
 927          */
 928         u64_stats_update_begin(&rx_stats->syncp);
 929         rx_stats->packets++;
 930         rx_stats->bytes += nvchan->rsc.pktlen;
 931
 932         if (skb->pkt_type == PACKET_BROADCAST)
 933                 ++rx_stats->broadcast;
 934         else if (skb->pkt_type == PACKET_MULTICAST)
 935                 ++rx_stats->multicast;
 936         u64_stats_update_end(&rx_stats->syncp);
 937
 938         if (act == XDP_TX) {
 939                 netvsc_xdp_xmit(skb, net);
 940                 return NVSP_STAT_SUCCESS;
 941         }
 942
 943         napi_gro_receive(&nvchan->napi, skb);
 944         return NVSP_STAT_SUCCESS;
 945 }
 946
 947 static void netvsc_get_drvinfo(struct net_device *net,
 948                                struct ethtool_drvinfo *info)
 949 {
 950         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
 951         strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
 952 }
 953
 954 static void netvsc_get_channels(struct net_device *net,
 955                                 struct ethtool_channels *channel)
 956 {
 957         struct net_device_context *net_device_ctx = netdev_priv(net);
 958         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
 959
 960         if (nvdev) {
 961                 channel->max_combined   = nvdev->max_chn;
 962                 channel->combined_count = nvdev->num_chn;
 963         }
 964 }
 965
 966 /* Alloc struct netvsc_device_info, and initialize it from either existing
 967  * struct netvsc_device, or from default values.
 968  */
 969 static
 970 struct netvsc_device_info *netvsc_devinfo_get(struct netvsc_device *nvdev)
 971 {
 972         struct netvsc_device_info *dev_info;
 973         struct bpf_prog *prog;
 974
 975         dev_info = kzalloc(sizeof(*dev_info), GFP_ATOMIC);
 976
 977         if (!dev_info)
 978                 return NULL;
 979
 980         if (nvdev) {
 981                 ASSERT_RTNL();
 982
 983                 dev_info->num_chn = nvdev->num_chn;
 984                 dev_info->send_sections = nvdev->send_section_cnt;
 985                 dev_info->send_section_size = nvdev->send_section_size;
 986                 dev_info->recv_sections = nvdev->recv_section_cnt;
 987                 dev_info->recv_section_size = nvdev->recv_section_size;
 988
 989                 memcpy(dev_info->rss_key, nvdev->extension->rss_key,
 990                        NETVSC_HASH_KEYLEN);
 991
 992                 prog = netvsc_xdp_get(nvdev);
 993                 if (prog) {
 994                         bpf_prog_inc(prog);
 995                         dev_info->bprog = prog;
 996                 }
 997         } else {
 998                 dev_info->num_chn = VRSS_CHANNEL_DEFAULT;
 999                 dev_info->send_sections = NETVSC_DEFAULT_TX;
1000                 dev_info->send_section_size = NETVSC_SEND_SECTION_SIZE;
1001                 dev_info->recv_sections = NETVSC_DEFAULT_RX;
1002                 dev_info->recv_section_size = NETVSC_RECV_SECTION_SIZE;
1003         }
1004
1005         return dev_info;
1006 }
1007
1008 /* Free struct netvsc_device_info */
1009 static void netvsc_devinfo_put(struct netvsc_device_info *dev_info)
1010 {
1011         if (dev_info->bprog) {
1012                 ASSERT_RTNL();
1013                 bpf_prog_put(dev_info->bprog);
1014         }
1015
1016         kfree(dev_info);
1017 }
1018
1019 static int netvsc_detach(struct net_device *ndev,
1020                          struct netvsc_device *nvdev)
1021 {
1022         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1023         struct hv_device *hdev = ndev_ctx->device_ctx;
1024         int ret;
1025
1026         /* Don't try continuing to try and setup sub channels */
1027         if (cancel_work_sync(&nvdev->subchan_work))
1028                 nvdev->num_chn = 1;
1029
1030         netvsc_xdp_set(ndev, NULL, NULL, nvdev);
1031
1032         /* If device was up (receiving) then shutdown */
1033         if (netif_running(ndev)) {
1034                 netvsc_tx_disable(nvdev, ndev);
1035
1036                 ret = rndis_filter_close(nvdev);
1037                 if (ret) {
1038                         netdev_err(ndev,
1039                                    "unable to close device (ret %d).\n", ret);
1040                         return ret;
1041                 }
1042
1043                 ret = netvsc_wait_until_empty(nvdev);
1044                 if (ret) {
1045                         netdev_err(ndev,
1046                                    "Ring buffer not empty after closing rndis\n");
1047                         return ret;
1048                 }
1049         }
1050
1051         netif_device_detach(ndev);
1052
1053         rndis_filter_device_remove(hdev, nvdev);
1054
1055         return 0;
1056 }
1057
1058 static int netvsc_attach(struct net_device *ndev,
1059                          struct netvsc_device_info *dev_info)
1060 {
1061         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1062         struct hv_device *hdev = ndev_ctx->device_ctx;
1063         struct netvsc_device *nvdev;
1064         struct rndis_device *rdev;
1065         struct bpf_prog *prog;
1066         int ret = 0;
1067
1068         nvdev = rndis_filter_device_add(hdev, dev_info);
1069         if (IS_ERR(nvdev))
1070                 return PTR_ERR(nvdev);
1071
1072         if (nvdev->num_chn > 1) {
1073                 ret = rndis_set_subchannel(ndev, nvdev, dev_info);
1074
1075                 /* if unavailable, just proceed with one queue */
1076                 if (ret) {
1077                         nvdev->max_chn = 1;
1078                         nvdev->num_chn = 1;
1079                 }
1080         }
1081
1082         prog = dev_info->bprog;
1083         if (prog) {
1084                 bpf_prog_inc(prog);
1085                 ret = netvsc_xdp_set(ndev, prog, NULL, nvdev);
1086                 if (ret) {
1087                         bpf_prog_put(prog);
1088                         goto err1;
1089                 }
1090         }
1091
1092         /* In any case device is now ready */
1093         nvdev->tx_disable = false;
1094         netif_device_attach(ndev);
1095
1096         /* Note: enable and attach happen when sub-channels setup */
1097         netif_carrier_off(ndev);
1098
1099         if (netif_running(ndev)) {
1100                 ret = rndis_filter_open(nvdev);
1101                 if (ret)
1102                         goto err2;
1103
1104                 rdev = nvdev->extension;
1105                 if (!rdev->link_state)
1106                         netif_carrier_on(ndev);
1107         }
1108
1109         return 0;
1110
1111 err2:
1112         netif_device_detach(ndev);
1113
1114 err1:
1115         rndis_filter_device_remove(hdev, nvdev);
1116
1117         return ret;
1118 }
1119
1120 static int netvsc_set_channels(struct net_device *net,
1121                                struct ethtool_channels *channels)
1122 {
1123         struct net_device_context *net_device_ctx = netdev_priv(net);
1124         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
1125         unsigned int orig, count = channels->combined_count;
1126         struct netvsc_device_info *device_info;
1127         int ret;
1128
1129         /* We do not support separate count for rx, tx, or other */
1130         if (count == 0 ||
1131             channels->rx_count || channels->tx_count || channels->other_count)
1132                 return -EINVAL;
1133
1134         if (!nvdev || nvdev->destroy)
1135                 return -ENODEV;
1136
1137         if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
1138                 return -EINVAL;
1139
1140         if (count > nvdev->max_chn)
1141                 return -EINVAL;
1142
1143         orig = nvdev->num_chn;
1144
1145         device_info = netvsc_devinfo_get(nvdev);
1146
1147         if (!device_info)
1148                 return -ENOMEM;
1149
1150         device_info->num_chn = count;
1151
1152         ret = netvsc_detach(net, nvdev);
1153         if (ret)
1154                 goto out;
1155
1156         ret = netvsc_attach(net, device_info);
1157         if (ret) {
1158                 device_info->num_chn = orig;
1159                 if (netvsc_attach(net, device_info))
1160                         netdev_err(net, "restoring channel setting failed\n");
1161         }
1162
1163 out:
1164         netvsc_devinfo_put(device_info);
1165         return ret;
1166 }
1167
1168 static void netvsc_init_settings(struct net_device *dev)
1169 {
1170         struct net_device_context *ndc = netdev_priv(dev);
1171
1172         ndc->l4_hash = HV_DEFAULT_L4HASH;
1173
1174         ndc->speed = SPEED_UNKNOWN;
1175         ndc->duplex = DUPLEX_FULL;
1176
1177         dev->features = NETIF_F_LRO;
1178 }
1179
1180 static int netvsc_get_link_ksettings(struct net_device *dev,
1181                                      struct ethtool_link_ksettings *cmd)
1182 {
1183         struct net_device_context *ndc = netdev_priv(dev);
1184         struct net_device *vf_netdev;
1185
1186         vf_netdev = rtnl_dereference(ndc->vf_netdev);
1187
1188         if (vf_netdev)
1189                 return __ethtool_get_link_ksettings(vf_netdev, cmd);
1190
1191         cmd->base.speed = ndc->speed;
1192         cmd->base.duplex = ndc->duplex;
1193         cmd->base.port = PORT_OTHER;
1194
1195         return 0;
1196 }
1197
1198 static int netvsc_set_link_ksettings(struct net_device *dev,
1199                                      const struct ethtool_link_ksettings *cmd)
1200 {
1201         struct net_device_context *ndc = netdev_priv(dev);
1202         struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1203
1204         if (vf_netdev) {
1205                 if (!vf_netdev->ethtool_ops->set_link_ksettings)
1206                         return -EOPNOTSUPP;
1207
1208                 return vf_netdev->ethtool_ops->set_link_ksettings(vf_netdev,
1209                                                                   cmd);
1210         }
1211
1212         return ethtool_virtdev_set_link_ksettings(dev, cmd,
1213                                                   &ndc->speed, &ndc->duplex);
1214 }
1215
1216 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
1217 {
1218         struct net_device_context *ndevctx = netdev_priv(ndev);
1219         struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1220         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1221         int orig_mtu = ndev->mtu;
1222         struct netvsc_device_info *device_info;
1223         int ret = 0;
1224
1225         if (!nvdev || nvdev->destroy)
1226                 return -ENODEV;
1227
1228         device_info = netvsc_devinfo_get(nvdev);
1229
1230         if (!device_info)
1231                 return -ENOMEM;
1232
1233         /* Change MTU of underlying VF netdev first. */
1234         if (vf_netdev) {
1235                 ret = dev_set_mtu(vf_netdev, mtu);
1236                 if (ret)
1237                         goto out;
1238         }
1239
1240         ret = netvsc_detach(ndev, nvdev);
1241         if (ret)
1242                 goto rollback_vf;
1243
1244         ndev->mtu = mtu;
1245
1246         ret = netvsc_attach(ndev, device_info);
1247         if (!ret)
1248                 goto out;
1249
1250         /* Attempt rollback to original MTU */
1251         ndev->mtu = orig_mtu;
1252
1253         if (netvsc_attach(ndev, device_info))
1254                 netdev_err(ndev, "restoring mtu failed\n");
1255 rollback_vf:
1256         if (vf_netdev)
1257                 dev_set_mtu(vf_netdev, orig_mtu);
1258
1259 out:
1260         netvsc_devinfo_put(device_info);
1261         return ret;
1262 }
1263
1264 static void netvsc_get_vf_stats(struct net_device *net,
1265                                 struct netvsc_vf_pcpu_stats *tot)
1266 {
1267         struct net_device_context *ndev_ctx = netdev_priv(net);
1268         int i;
1269
1270         memset(tot, 0, sizeof(*tot));
1271
1272         for_each_possible_cpu(i) {
1273                 const struct netvsc_vf_pcpu_stats *stats
1274                         = per_cpu_ptr(ndev_ctx->vf_stats, i);
1275                 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1276                 unsigned int start;
1277
1278                 do {
1279                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1280                         rx_packets = stats->rx_packets;
1281                         tx_packets = stats->tx_packets;
1282                         rx_bytes = stats->rx_bytes;
1283                         tx_bytes = stats->tx_bytes;
1284                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1285
1286                 tot->rx_packets += rx_packets;
1287                 tot->tx_packets += tx_packets;
1288                 tot->rx_bytes   += rx_bytes;
1289                 tot->tx_bytes   += tx_bytes;
1290                 tot->tx_dropped += stats->tx_dropped;
1291         }
1292 }
1293
1294 static void netvsc_get_pcpu_stats(struct net_device *net,
1295                                   struct netvsc_ethtool_pcpu_stats *pcpu_tot)
1296 {
1297         struct net_device_context *ndev_ctx = netdev_priv(net);
1298         struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
1299         int i;
1300
1301         /* fetch percpu stats of vf */
1302         for_each_possible_cpu(i) {
1303                 const struct netvsc_vf_pcpu_stats *stats =
1304                         per_cpu_ptr(ndev_ctx->vf_stats, i);
1305                 struct netvsc_ethtool_pcpu_stats *this_tot = &pcpu_tot[i];
1306                 unsigned int start;
1307
1308                 do {
1309                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1310                         this_tot->vf_rx_packets = stats->rx_packets;
1311                         this_tot->vf_tx_packets = stats->tx_packets;
1312                         this_tot->vf_rx_bytes = stats->rx_bytes;
1313                         this_tot->vf_tx_bytes = stats->tx_bytes;
1314                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1315                 this_tot->rx_packets = this_tot->vf_rx_packets;
1316                 this_tot->tx_packets = this_tot->vf_tx_packets;
1317                 this_tot->rx_bytes   = this_tot->vf_rx_bytes;
1318                 this_tot->tx_bytes   = this_tot->vf_tx_bytes;
1319         }
1320
1321         /* fetch percpu stats of netvsc */
1322         for (i = 0; i < nvdev->num_chn; i++) {
1323                 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1324                 const struct netvsc_stats *stats;
1325                 struct netvsc_ethtool_pcpu_stats *this_tot =
1326                         &pcpu_tot[nvchan->channel->target_cpu];
1327                 u64 packets, bytes;
1328                 unsigned int start;
1329
1330                 stats = &nvchan->tx_stats;
1331                 do {
1332                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1333                         packets = stats->packets;
1334                         bytes = stats->bytes;
1335                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1336
1337                 this_tot->tx_bytes      += bytes;
1338                 this_tot->tx_packets    += packets;
1339
1340                 stats = &nvchan->rx_stats;
1341                 do {
1342                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1343                         packets = stats->packets;
1344                         bytes = stats->bytes;
1345                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1346
1347                 this_tot->rx_bytes      += bytes;
1348                 this_tot->rx_packets    += packets;
1349         }
1350 }
1351
1352 static void netvsc_get_stats64(struct net_device *net,
1353                                struct rtnl_link_stats64 *t)
1354 {
1355         struct net_device_context *ndev_ctx = netdev_priv(net);
1356         struct netvsc_device *nvdev;
1357         struct netvsc_vf_pcpu_stats vf_tot;
1358         int i;
1359
1360         rcu_read_lock();
1361
1362         nvdev = rcu_dereference(ndev_ctx->nvdev);
1363         if (!nvdev)
1364                 goto out;
1365
1366         netdev_stats_to_stats64(t, &net->stats);
1367
1368         netvsc_get_vf_stats(net, &vf_tot);
1369         t->rx_packets += vf_tot.rx_packets;
1370         t->tx_packets += vf_tot.tx_packets;
1371         t->rx_bytes   += vf_tot.rx_bytes;
1372         t->tx_bytes   += vf_tot.tx_bytes;
1373         t->tx_dropped += vf_tot.tx_dropped;
1374
1375         for (i = 0; i < nvdev->num_chn; i++) {
1376                 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1377                 const struct netvsc_stats *stats;
1378                 u64 packets, bytes, multicast;
1379                 unsigned int start;
1380
1381                 stats = &nvchan->tx_stats;
1382                 do {
1383                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1384                         packets = stats->packets;
1385                         bytes = stats->bytes;
1386                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1387
1388                 t->tx_bytes     += bytes;
1389                 t->tx_packets   += packets;
1390
1391                 stats = &nvchan->rx_stats;
1392                 do {
1393                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1394                         packets = stats->packets;
1395                         bytes = stats->bytes;
1396                         multicast = stats->multicast + stats->broadcast;
1397                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1398
1399                 t->rx_bytes     += bytes;
1400                 t->rx_packets   += packets;
1401                 t->multicast    += multicast;
1402         }
1403 out:
1404         rcu_read_unlock();
1405 }
1406
1407 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
1408 {
1409         struct net_device_context *ndc = netdev_priv(ndev);
1410         struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1411         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1412         struct sockaddr *addr = p;
1413         int err;
1414
1415         err = eth_prepare_mac_addr_change(ndev, p);
1416         if (err)
1417                 return err;
1418
1419         if (!nvdev)
1420                 return -ENODEV;
1421
1422         if (vf_netdev) {
1423                 err = dev_set_mac_address(vf_netdev, addr, NULL);
1424                 if (err)
1425                         return err;
1426         }
1427
1428         err = rndis_filter_set_device_mac(nvdev, addr->sa_data);
1429         if (!err) {
1430                 eth_commit_mac_addr_change(ndev, p);
1431         } else if (vf_netdev) {
1432                 /* rollback change on VF */
1433                 memcpy(addr->sa_data, ndev->dev_addr, ETH_ALEN);
1434                 dev_set_mac_address(vf_netdev, addr, NULL);
1435         }
1436
1437         return err;
1438 }
1439
1440 static const struct {
1441         char name[ETH_GSTRING_LEN];
1442         u16 offset;
1443 } netvsc_stats[] = {
1444         { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
1445         { "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
1446         { "tx_no_space",  offsetof(struct netvsc_ethtool_stats, tx_no_space) },
1447         { "tx_too_big",   offsetof(struct netvsc_ethtool_stats, tx_too_big) },
1448         { "tx_busy",      offsetof(struct netvsc_ethtool_stats, tx_busy) },
1449         { "tx_send_full", offsetof(struct netvsc_ethtool_stats, tx_send_full) },
1450         { "rx_comp_busy", offsetof(struct netvsc_ethtool_stats, rx_comp_busy) },
1451         { "rx_no_memory", offsetof(struct netvsc_ethtool_stats, rx_no_memory) },
1452         { "stop_queue", offsetof(struct netvsc_ethtool_stats, stop_queue) },
1453         { "wake_queue", offsetof(struct netvsc_ethtool_stats, wake_queue) },
1454         { "vlan_error", offsetof(struct netvsc_ethtool_stats, vlan_error) },
1455 }, pcpu_stats[] = {
1456         { "cpu%u_rx_packets",
1457                 offsetof(struct netvsc_ethtool_pcpu_stats, rx_packets) },
1458         { "cpu%u_rx_bytes",
1459                 offsetof(struct netvsc_ethtool_pcpu_stats, rx_bytes) },
1460         { "cpu%u_tx_packets",
1461                 offsetof(struct netvsc_ethtool_pcpu_stats, tx_packets) },
1462         { "cpu%u_tx_bytes",
1463                 offsetof(struct netvsc_ethtool_pcpu_stats, tx_bytes) },
1464         { "cpu%u_vf_rx_packets",
1465                 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_packets) },
1466         { "cpu%u_vf_rx_bytes",
1467                 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_bytes) },
1468         { "cpu%u_vf_tx_packets",
1469                 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_packets) },
1470         { "cpu%u_vf_tx_bytes",
1471                 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_bytes) },
1472 }, vf_stats[] = {
1473         { "vf_rx_packets", offsetof(struct netvsc_vf_pcpu_stats, rx_packets) },
1474         { "vf_rx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, rx_bytes) },
1475         { "vf_tx_packets", offsetof(struct netvsc_vf_pcpu_stats, tx_packets) },
1476         { "vf_tx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, tx_bytes) },
1477         { "vf_tx_dropped", offsetof(struct netvsc_vf_pcpu_stats, tx_dropped) },
1478 };
1479
1480 #define NETVSC_GLOBAL_STATS_LEN ARRAY_SIZE(netvsc_stats)
1481 #define NETVSC_VF_STATS_LEN     ARRAY_SIZE(vf_stats)
1482
1483 /* statistics per queue (rx/tx packets/bytes) */
1484 #define NETVSC_PCPU_STATS_LEN (num_present_cpus() * ARRAY_SIZE(pcpu_stats))
1485
1486 /* 5 statistics per queue (rx/tx packets/bytes, rx xdp_drop) */
1487 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 5)
1488
1489 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
1490 {
1491         struct net_device_context *ndc = netdev_priv(dev);
1492         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1493
1494         if (!nvdev)
1495                 return -ENODEV;
1496
1497         switch (string_set) {
1498         case ETH_SS_STATS:
1499                 return NETVSC_GLOBAL_STATS_LEN
1500                         + NETVSC_VF_STATS_LEN
1501                         + NETVSC_QUEUE_STATS_LEN(nvdev)
1502                         + NETVSC_PCPU_STATS_LEN;
1503         default:
1504                 return -EINVAL;
1505         }
1506 }
1507
1508 static void netvsc_get_ethtool_stats(struct net_device *dev,
1509                                      struct ethtool_stats *stats, u64 *data)
1510 {
1511         struct net_device_context *ndc = netdev_priv(dev);
1512         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1513         const void *nds = &ndc->eth_stats;
1514         const struct netvsc_stats *qstats;
1515         struct netvsc_vf_pcpu_stats sum;
1516         struct netvsc_ethtool_pcpu_stats *pcpu_sum;
1517         unsigned int start;
1518         u64 packets, bytes;
1519         u64 xdp_drop;
1520         int i, j, cpu;
1521
1522         if (!nvdev)
1523                 return;
1524
1525         for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
1526                 data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1527
1528         netvsc_get_vf_stats(dev, &sum);
1529         for (j = 0; j < NETVSC_VF_STATS_LEN; j++)
1530                 data[i++] = *(u64 *)((void *)&sum + vf_stats[j].offset);
1531
1532         for (j = 0; j < nvdev->num_chn; j++) {
1533                 qstats = &nvdev->chan_table[j].tx_stats;
1534
1535                 do {
1536                         start = u64_stats_fetch_begin_irq(&qstats->syncp);
1537                         packets = qstats->packets;
1538                         bytes = qstats->bytes;
1539                 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1540                 data[i++] = packets;
1541                 data[i++] = bytes;
1542
1543                 qstats = &nvdev->chan_table[j].rx_stats;
1544                 do {
1545                         start = u64_stats_fetch_begin_irq(&qstats->syncp);
1546                         packets = qstats->packets;
1547                         bytes = qstats->bytes;
1548                         xdp_drop = qstats->xdp_drop;
1549                 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1550                 data[i++] = packets;
1551                 data[i++] = bytes;
1552                 data[i++] = xdp_drop;
1553         }
1554
1555         pcpu_sum = kvmalloc_array(num_possible_cpus(),
1556                                   sizeof(struct netvsc_ethtool_pcpu_stats),
1557                                   GFP_KERNEL);
1558         netvsc_get_pcpu_stats(dev, pcpu_sum);
1559         for_each_present_cpu(cpu) {
1560                 struct netvsc_ethtool_pcpu_stats *this_sum = &pcpu_sum[cpu];
1561
1562                 for (j = 0; j < ARRAY_SIZE(pcpu_stats); j++)
1563                         data[i++] = *(u64 *)((void *)this_sum
1564                                              + pcpu_stats[j].offset);
1565         }
1566         kvfree(pcpu_sum);
1567 }
1568
1569 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1570 {
1571         struct net_device_context *ndc = netdev_priv(dev);
1572         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1573         u8 *p = data;
1574         int i, cpu;
1575
1576         if (!nvdev)
1577                 return;
1578
1579         switch (stringset) {
1580         case ETH_SS_STATS:
1581                 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++) {
1582                         memcpy(p, netvsc_stats[i].name, ETH_GSTRING_LEN);
1583                         p += ETH_GSTRING_LEN;
1584                 }
1585
1586                 for (i = 0; i < ARRAY_SIZE(vf_stats); i++) {
1587                         memcpy(p, vf_stats[i].name, ETH_GSTRING_LEN);
1588                         p += ETH_GSTRING_LEN;
1589                 }
1590
1591                 for (i = 0; i < nvdev->num_chn; i++) {
1592                         sprintf(p, "tx_queue_%u_packets", i);
1593                         p += ETH_GSTRING_LEN;
1594                         sprintf(p, "tx_queue_%u_bytes", i);
1595                         p += ETH_GSTRING_LEN;
1596                         sprintf(p, "rx_queue_%u_packets", i);
1597                         p += ETH_GSTRING_LEN;
1598                         sprintf(p, "rx_queue_%u_bytes", i);
1599                         p += ETH_GSTRING_LEN;
1600                         sprintf(p, "rx_queue_%u_xdp_drop", i);
1601                         p += ETH_GSTRING_LEN;
1602                 }
1603
1604                 for_each_present_cpu(cpu) {
1605                         for (i = 0; i < ARRAY_SIZE(pcpu_stats); i++) {
1606                                 sprintf(p, pcpu_stats[i].name, cpu);
1607                                 p += ETH_GSTRING_LEN;
1608                         }
1609                 }
1610
1611                 break;
1612         }
1613 }
1614
1615 static int
1616 netvsc_get_rss_hash_opts(struct net_device_context *ndc,
1617                          struct ethtool_rxnfc *info)
1618 {
1619         const u32 l4_flag = RXH_L4_B_0_1 | RXH_L4_B_2_3;
1620
1621         info->data = RXH_IP_SRC | RXH_IP_DST;
1622
1623         switch (info->flow_type) {
1624         case TCP_V4_FLOW:
1625                 if (ndc->l4_hash & HV_TCP4_L4HASH)
1626                         info->data |= l4_flag;
1627
1628                 break;
1629
1630         case TCP_V6_FLOW:
1631                 if (ndc->l4_hash & HV_TCP6_L4HASH)
1632                         info->data |= l4_flag;
1633
1634                 break;
1635
1636         case UDP_V4_FLOW:
1637                 if (ndc->l4_hash & HV_UDP4_L4HASH)
1638                         info->data |= l4_flag;
1639
1640                 break;
1641
1642         case UDP_V6_FLOW:
1643                 if (ndc->l4_hash & HV_UDP6_L4HASH)
1644                         info->data |= l4_flag;
1645
1646                 break;
1647
1648         case IPV4_FLOW:
1649         case IPV6_FLOW:
1650                 break;
1651         default:
1652                 info->data = 0;
1653                 break;
1654         }
1655
1656         return 0;
1657 }
1658
1659 static int
1660 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1661                  u32 *rules)
1662 {
1663         struct net_device_context *ndc = netdev_priv(dev);
1664         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1665
1666         if (!nvdev)
1667                 return -ENODEV;
1668
1669         switch (info->cmd) {
1670         case ETHTOOL_GRXRINGS:
1671                 info->data = nvdev->num_chn;
1672                 return 0;
1673
1674         case ETHTOOL_GRXFH:
1675                 return netvsc_get_rss_hash_opts(ndc, info);
1676         }
1677         return -EOPNOTSUPP;
1678 }
1679
1680 static int netvsc_set_rss_hash_opts(struct net_device_context *ndc,
1681                                     struct ethtool_rxnfc *info)
1682 {
1683         if (info->data == (RXH_IP_SRC | RXH_IP_DST |
1684                            RXH_L4_B_0_1 | RXH_L4_B_2_3)) {
1685                 switch (info->flow_type) {
1686                 case TCP_V4_FLOW:
1687                         ndc->l4_hash |= HV_TCP4_L4HASH;
1688                         break;
1689
1690                 case TCP_V6_FLOW:
1691                         ndc->l4_hash |= HV_TCP6_L4HASH;
1692                         break;
1693
1694                 case UDP_V4_FLOW:
1695                         ndc->l4_hash |= HV_UDP4_L4HASH;
1696                         break;
1697
1698                 case UDP_V6_FLOW:
1699                         ndc->l4_hash |= HV_UDP6_L4HASH;
1700                         break;
1701
1702                 default:
1703                         return -EOPNOTSUPP;
1704                 }
1705
1706                 return 0;
1707         }
1708
1709         if (info->data == (RXH_IP_SRC | RXH_IP_DST)) {
1710                 switch (info->flow_type) {
1711                 case TCP_V4_FLOW:
1712                         ndc->l4_hash &= ~HV_TCP4_L4HASH;
1713                         break;
1714
1715                 case TCP_V6_FLOW:
1716                         ndc->l4_hash &= ~HV_TCP6_L4HASH;
1717                         break;
1718
1719                 case UDP_V4_FLOW:
1720                         ndc->l4_hash &= ~HV_UDP4_L4HASH;
1721                         break;
1722
1723                 case UDP_V6_FLOW:
1724                         ndc->l4_hash &= ~HV_UDP6_L4HASH;
1725                         break;
1726
1727                 default:
1728                         return -EOPNOTSUPP;
1729                 }
1730
1731                 return 0;
1732         }
1733
1734         return -EOPNOTSUPP;
1735 }
1736
1737 static int
1738 netvsc_set_rxnfc(struct net_device *ndev, struct ethtool_rxnfc *info)
1739 {
1740         struct net_device_context *ndc = netdev_priv(ndev);
1741
1742         if (info->cmd == ETHTOOL_SRXFH)
1743                 return netvsc_set_rss_hash_opts(ndc, info);
1744
1745         return -EOPNOTSUPP;
1746 }
1747
1748 static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1749 {
1750         return NETVSC_HASH_KEYLEN;
1751 }
1752
1753 static u32 netvsc_rss_indir_size(struct net_device *dev)
1754 {
1755         return ITAB_NUM;
1756 }
1757
1758 static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key,
1759                            u8 *hfunc)
1760 {
1761         struct net_device_context *ndc = netdev_priv(dev);
1762         struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1763         struct rndis_device *rndis_dev;
1764         int i;
1765
1766         if (!ndev)
1767                 return -ENODEV;
1768
1769         if (hfunc)
1770                 *hfunc = ETH_RSS_HASH_TOP;      /* Toeplitz */
1771
1772         rndis_dev = ndev->extension;
1773         if (indir) {
1774                 for (i = 0; i < ITAB_NUM; i++)
1775                         indir[i] = ndc->rx_table[i];
1776         }
1777
1778         if (key)
1779                 memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1780
1781         return 0;
1782 }
1783
1784 static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
1785                            const u8 *key, const u8 hfunc)
1786 {
1787         struct net_device_context *ndc = netdev_priv(dev);
1788         struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1789         struct rndis_device *rndis_dev;
1790         int i;
1791
1792         if (!ndev)
1793                 return -ENODEV;
1794
1795         if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
1796                 return -EOPNOTSUPP;
1797
1798         rndis_dev = ndev->extension;
1799         if (indir) {
1800                 for (i = 0; i < ITAB_NUM; i++)
1801                         if (indir[i] >= ndev->num_chn)
1802                                 return -EINVAL;
1803
1804                 for (i = 0; i < ITAB_NUM; i++)
1805                         ndc->rx_table[i] = indir[i];
1806         }
1807
1808         if (!key) {
1809                 if (!indir)
1810                         return 0;
1811
1812                 key = rndis_dev->rss_key;
1813         }
1814
1815         return rndis_filter_set_rss_param(rndis_dev, key);
1816 }
1817
1818 /* Hyper-V RNDIS protocol does not have ring in the HW sense.
1819  * It does have pre-allocated receive area which is divided into sections.
1820  */
1821 static void __netvsc_get_ringparam(struct netvsc_device *nvdev,
1822                                    struct ethtool_ringparam *ring)
1823 {
1824         u32 max_buf_size;
1825
1826         ring->rx_pending = nvdev->recv_section_cnt;
1827         ring->tx_pending = nvdev->send_section_cnt;
1828
1829         if (nvdev->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
1830                 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
1831         else
1832                 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
1833
1834         ring->rx_max_pending = max_buf_size / nvdev->recv_section_size;
1835         ring->tx_max_pending = NETVSC_SEND_BUFFER_SIZE
1836                 / nvdev->send_section_size;
1837 }
1838
1839 static void netvsc_get_ringparam(struct net_device *ndev,
1840                                  struct ethtool_ringparam *ring)
1841 {
1842         struct net_device_context *ndevctx = netdev_priv(ndev);
1843         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1844
1845         if (!nvdev)
1846                 return;
1847
1848         __netvsc_get_ringparam(nvdev, ring);
1849 }
1850
1851 static int netvsc_set_ringparam(struct net_device *ndev,
1852                                 struct ethtool_ringparam *ring)
1853 {
1854         struct net_device_context *ndevctx = netdev_priv(ndev);
1855         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1856         struct netvsc_device_info *device_info;
1857         struct ethtool_ringparam orig;
1858         u32 new_tx, new_rx;
1859         int ret = 0;
1860
1861         if (!nvdev || nvdev->destroy)
1862                 return -ENODEV;
1863
1864         memset(&orig, 0, sizeof(orig));
1865         __netvsc_get_ringparam(nvdev, &orig);
1866
1867         new_tx = clamp_t(u32, ring->tx_pending,
1868                          NETVSC_MIN_TX_SECTIONS, orig.tx_max_pending);
1869         new_rx = clamp_t(u32, ring->rx_pending,
1870                          NETVSC_MIN_RX_SECTIONS, orig.rx_max_pending);
1871
1872         if (new_tx == orig.tx_pending &&
1873             new_rx == orig.rx_pending)
1874                 return 0;        /* no change */
1875
1876         device_info = netvsc_devinfo_get(nvdev);
1877
1878         if (!device_info)
1879                 return -ENOMEM;
1880
1881         device_info->send_sections = new_tx;
1882         device_info->recv_sections = new_rx;
1883
1884         ret = netvsc_detach(ndev, nvdev);
1885         if (ret)
1886                 goto out;
1887
1888         ret = netvsc_attach(ndev, device_info);
1889         if (ret) {
1890                 device_info->send_sections = orig.tx_pending;
1891                 device_info->recv_sections = orig.rx_pending;
1892
1893                 if (netvsc_attach(ndev, device_info))
1894                         netdev_err(ndev, "restoring ringparam failed");
1895         }
1896
1897 out:
1898         netvsc_devinfo_put(device_info);
1899         return ret;
1900 }
1901
1902 static netdev_features_t netvsc_fix_features(struct net_device *ndev,
1903                                              netdev_features_t features)
1904 {
1905         struct net_device_context *ndevctx = netdev_priv(ndev);
1906         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1907
1908         if (!nvdev || nvdev->destroy)
1909                 return features;
1910
1911         if ((features & NETIF_F_LRO) && netvsc_xdp_get(nvdev)) {
1912                 features ^= NETIF_F_LRO;
1913                 netdev_info(ndev, "Skip LRO - unsupported with XDP\n");
1914         }
1915
1916         return features;
1917 }
1918
1919 static int netvsc_set_features(struct net_device *ndev,
1920                                netdev_features_t features)
1921 {
1922         netdev_features_t change = features ^ ndev->features;
1923         struct net_device_context *ndevctx = netdev_priv(ndev);
1924         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1925         struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1926         struct ndis_offload_params offloads;
1927         int ret = 0;
1928
1929         if (!nvdev || nvdev->destroy)
1930                 return -ENODEV;
1931
1932         if (!(change & NETIF_F_LRO))
1933                 goto syncvf;
1934
1935         memset(&offloads, 0, sizeof(struct ndis_offload_params));
1936
1937         if (features & NETIF_F_LRO) {
1938                 offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1939                 offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1940         } else {
1941                 offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1942                 offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1943         }
1944
1945         ret = rndis_filter_set_offload_params(ndev, nvdev, &offloads);
1946
1947         if (ret) {
1948                 features ^= NETIF_F_LRO;
1949                 ndev->features = features;
1950         }
1951
1952 syncvf:
1953         if (!vf_netdev)
1954                 return ret;
1955
1956         vf_netdev->wanted_features = features;
1957         netdev_update_features(vf_netdev);
1958
1959         return ret;
1960 }
1961
1962 static int netvsc_get_regs_len(struct net_device *netdev)
1963 {
1964         return VRSS_SEND_TAB_SIZE * sizeof(u32);
1965 }
1966
1967 static void netvsc_get_regs(struct net_device *netdev,
1968                             struct ethtool_regs *regs, void *p)
1969 {
1970         struct net_device_context *ndc = netdev_priv(netdev);
1971         u32 *regs_buff = p;
1972
1973         /* increase the version, if buffer format is changed. */
1974         regs->version = 1;
1975
1976         memcpy(regs_buff, ndc->tx_table, VRSS_SEND_TAB_SIZE * sizeof(u32));
1977 }
1978
1979 static u32 netvsc_get_msglevel(struct net_device *ndev)
1980 {
1981         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1982
1983         return ndev_ctx->msg_enable;
1984 }
1985
1986 static void netvsc_set_msglevel(struct net_device *ndev, u32 val)
1987 {
1988         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1989
1990         ndev_ctx->msg_enable = val;
1991 }
1992
1993 static const struct ethtool_ops ethtool_ops = {
1994         .get_drvinfo    = netvsc_get_drvinfo,
1995         .get_regs_len   = netvsc_get_regs_len,
1996         .get_regs       = netvsc_get_regs,
1997         .get_msglevel   = netvsc_get_msglevel,
1998         .set_msglevel   = netvsc_set_msglevel,
1999         .get_link       = ethtool_op_get_link,
2000         .get_ethtool_stats = netvsc_get_ethtool_stats,
2001         .get_sset_count = netvsc_get_sset_count,
2002         .get_strings    = netvsc_get_strings,
2003         .get_channels   = netvsc_get_channels,
2004         .set_channels   = netvsc_set_channels,
2005         .get_ts_info    = ethtool_op_get_ts_info,
2006         .get_rxnfc      = netvsc_get_rxnfc,
2007         .set_rxnfc      = netvsc_set_rxnfc,
2008         .get_rxfh_key_size = netvsc_get_rxfh_key_size,
2009         .get_rxfh_indir_size = netvsc_rss_indir_size,
2010         .get_rxfh       = netvsc_get_rxfh,
2011         .set_rxfh       = netvsc_set_rxfh,
2012         .get_link_ksettings = netvsc_get_link_ksettings,
2013         .set_link_ksettings = netvsc_set_link_ksettings,
2014         .get_ringparam  = netvsc_get_ringparam,
2015         .set_ringparam  = netvsc_set_ringparam,
2016 };
2017
2018 static const struct net_device_ops device_ops = {
2019         .ndo_open =                     netvsc_open,
2020         .ndo_stop =                     netvsc_close,
2021         .ndo_start_xmit =               netvsc_start_xmit,
2022         .ndo_change_rx_flags =          netvsc_change_rx_flags,
2023         .ndo_set_rx_mode =              netvsc_set_rx_mode,
2024         .ndo_fix_features =             netvsc_fix_features,
2025         .ndo_set_features =             netvsc_set_features,
2026         .ndo_change_mtu =               netvsc_change_mtu,
2027         .ndo_validate_addr =            eth_validate_addr,
2028         .ndo_set_mac_address =          netvsc_set_mac_addr,
2029         .ndo_select_queue =             netvsc_select_queue,
2030         .ndo_get_stats64 =              netvsc_get_stats64,
2031         .ndo_bpf =                      netvsc_bpf,
2032 };
2033
2034 /*
2035  * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
2036  * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
2037  * present send GARP packet to network peers with netif_notify_peers().
2038  */
2039 static void netvsc_link_change(struct work_struct *w)
2040 {
2041         struct net_device_context *ndev_ctx =
2042                 container_of(w, struct net_device_context, dwork.work);
2043         struct hv_device *device_obj = ndev_ctx->device_ctx;
2044         struct net_device *net = hv_get_drvdata(device_obj);
2045         struct netvsc_device *net_device;
2046         struct rndis_device *rdev;
2047         struct netvsc_reconfig *event = NULL;
2048         bool notify = false, reschedule = false;
2049         unsigned long flags, next_reconfig, delay;
2050
2051         /* if changes are happening, comeback later */
2052         if (!rtnl_trylock()) {
2053                 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
2054                 return;
2055         }
2056
2057         net_device = rtnl_dereference(ndev_ctx->nvdev);
2058         if (!net_device)
2059                 goto out_unlock;
2060
2061         rdev = net_device->extension;
2062
2063         next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
2064         if (time_is_after_jiffies(next_reconfig)) {
2065                 /* link_watch only sends one notification with current state
2066                  * per second, avoid doing reconfig more frequently. Handle
2067                  * wrap around.
2068                  */
2069                 delay = next_reconfig - jiffies;
2070                 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
2071                 schedule_delayed_work(&ndev_ctx->dwork, delay);
2072                 goto out_unlock;
2073         }
2074         ndev_ctx->last_reconfig = jiffies;
2075
2076         spin_lock_irqsave(&ndev_ctx->lock, flags);
2077         if (!list_empty(&ndev_ctx->reconfig_events)) {
2078                 event = list_first_entry(&ndev_ctx->reconfig_events,
2079                                          struct netvsc_reconfig, list);
2080                 list_del(&event->list);
2081                 reschedule = !list_empty(&ndev_ctx->reconfig_events);
2082         }
2083         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
2084
2085         if (!event)
2086                 goto out_unlock;
2087
2088         switch (event->event) {
2089                 /* Only the following events are possible due to the check in
2090                  * netvsc_linkstatus_callback()
2091                  */
2092         case RNDIS_STATUS_MEDIA_CONNECT:
2093                 if (rdev->link_state) {
2094                         rdev->link_state = false;
2095                         netif_carrier_on(net);
2096                         netvsc_tx_enable(net_device, net);
2097                 } else {
2098                         notify = true;
2099                 }
2100                 kfree(event);
2101                 break;
2102         case RNDIS_STATUS_MEDIA_DISCONNECT:
2103                 if (!rdev->link_state) {
2104                         rdev->link_state = true;
2105                         netif_carrier_off(net);
2106                         netvsc_tx_disable(net_device, net);
2107                 }
2108                 kfree(event);
2109                 break;
2110         case RNDIS_STATUS_NETWORK_CHANGE:
2111                 /* Only makes sense if carrier is present */
2112                 if (!rdev->link_state) {
2113                         rdev->link_state = true;
2114                         netif_carrier_off(net);
2115                         netvsc_tx_disable(net_device, net);
2116                         event->event = RNDIS_STATUS_MEDIA_CONNECT;
2117                         spin_lock_irqsave(&ndev_ctx->lock, flags);
2118                         list_add(&event->list, &ndev_ctx->reconfig_events);
2119                         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
2120                         reschedule = true;
2121                 }
2122                 break;
2123         }
2124
2125         rtnl_unlock();
2126
2127         if (notify)
2128                 netdev_notify_peers(net);
2129
2130         /* link_watch only sends one notification with current state per
2131          * second, handle next reconfig event in 2 seconds.
2132          */
2133         if (reschedule)
2134                 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
2135
2136         return;
2137
2138 out_unlock:
2139         rtnl_unlock();
2140 }
2141
2142 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
2143 {
2144         struct net_device_context *net_device_ctx;
2145         struct net_device *dev;
2146
2147         dev = netdev_master_upper_dev_get(vf_netdev);
2148         if (!dev || dev->netdev_ops != &device_ops)
2149                 return NULL;    /* not a netvsc device */
2150
2151         net_device_ctx = netdev_priv(dev);
2152         if (!rtnl_dereference(net_device_ctx->nvdev))
2153                 return NULL;    /* device is removed */
2154
2155         return dev;
2156 }
2157
2158 /* Called when VF is injecting data into network stack.
2159  * Change the associated network device from VF to netvsc.
2160  * note: already called with rcu_read_lock
2161  */
2162 static rx_handler_result_t netvsc_vf_handle_frame(struct sk_buff **pskb)
2163 {
2164         struct sk_buff *skb = *pskb;
2165         struct net_device *ndev = rcu_dereference(skb->dev->rx_handler_data);
2166         struct net_device_context *ndev_ctx = netdev_priv(ndev);
2167         struct netvsc_vf_pcpu_stats *pcpu_stats
2168                  = this_cpu_ptr(ndev_ctx->vf_stats);
2169
2170         skb = skb_share_check(skb, GFP_ATOMIC);
2171         if (unlikely(!skb))
2172                 return RX_HANDLER_CONSUMED;
2173
2174         *pskb = skb;
2175
2176         skb->dev = ndev;
2177
2178         u64_stats_update_begin(&pcpu_stats->syncp);
2179         pcpu_stats->rx_packets++;
2180         pcpu_stats->rx_bytes += skb->len;
2181         u64_stats_update_end(&pcpu_stats->syncp);
2182
2183         return RX_HANDLER_ANOTHER;
2184 }
2185
2186 static int netvsc_vf_join(struct net_device *vf_netdev,
2187                           struct net_device *ndev)
2188 {
2189         struct net_device_context *ndev_ctx = netdev_priv(ndev);
2190         int ret;
2191
2192         ret = netdev_rx_handler_register(vf_netdev,
2193                                          netvsc_vf_handle_frame, ndev);
2194         if (ret != 0) {
2195                 netdev_err(vf_netdev,
2196                            "can not register netvsc VF receive handler (err = %d)\n",
2197                            ret);
2198                 goto rx_handler_failed;
2199         }
2200
2201         ret = netdev_master_upper_dev_link(vf_netdev, ndev,
2202                                            NULL, NULL, NULL);
2203         if (ret != 0) {
2204                 netdev_err(vf_netdev,
2205                            "can not set master device %s (err = %d)\n",
2206                            ndev->name, ret);
2207                 goto upper_link_failed;
2208         }
2209
2210         /* set slave flag before open to prevent IPv6 addrconf */
2211         vf_netdev->flags |= IFF_SLAVE;
2212
2213         schedule_delayed_work(&ndev_ctx->vf_takeover, VF_TAKEOVER_INT);
2214
2215         call_netdevice_notifiers(NETDEV_JOIN, vf_netdev);
2216
2217         netdev_info(vf_netdev, "joined to %s\n", ndev->name);
2218         return 0;
2219
2220 upper_link_failed:
2221         netdev_rx_handler_unregister(vf_netdev);
2222 rx_handler_failed:
2223         return ret;
2224 }
2225
2226 static void __netvsc_vf_setup(struct net_device *ndev,
2227                               struct net_device *vf_netdev)
2228 {
2229         int ret;
2230
2231         /* Align MTU of VF with master */
2232         ret = dev_set_mtu(vf_netdev, ndev->mtu);
2233         if (ret)
2234                 netdev_warn(vf_netdev,
2235                             "unable to change mtu to %u\n", ndev->mtu);
2236
2237         /* set multicast etc flags on VF */
2238         dev_change_flags(vf_netdev, ndev->flags | IFF_SLAVE, NULL);
2239
2240         /* sync address list from ndev to VF */
2241         netif_addr_lock_bh(ndev);
2242         dev_uc_sync(vf_netdev, ndev);
2243         dev_mc_sync(vf_netdev, ndev);
2244         netif_addr_unlock_bh(ndev);
2245
2246         if (netif_running(ndev)) {
2247                 ret = dev_open(vf_netdev, NULL);
2248                 if (ret)
2249                         netdev_warn(vf_netdev,
2250                                     "unable to open: %d\n", ret);
2251         }
2252 }
2253
2254 /* Setup VF as slave of the synthetic device.
2255  * Runs in workqueue to avoid recursion in netlink callbacks.
2256  */
2257 static void netvsc_vf_setup(struct work_struct *w)
2258 {
2259         struct net_device_context *ndev_ctx
2260                 = container_of(w, struct net_device_context, vf_takeover.work);
2261         struct net_device *ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2262         struct net_device *vf_netdev;
2263
2264         if (!rtnl_trylock()) {
2265                 schedule_delayed_work(&ndev_ctx->vf_takeover, 0);
2266                 return;
2267         }
2268
2269         vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2270         if (vf_netdev)
2271                 __netvsc_vf_setup(ndev, vf_netdev);
2272
2273         rtnl_unlock();
2274 }
2275
2276 /* Find netvsc by VF serial number.
2277  * The PCI hyperv controller records the serial number as the slot kobj name.
2278  */
2279 static struct net_device *get_netvsc_byslot(const struct net_device *vf_netdev)
2280 {
2281         struct device *parent = vf_netdev->dev.parent;
2282         struct net_device_context *ndev_ctx;
2283         struct pci_dev *pdev;
2284         u32 serial;
2285
2286         if (!parent || !dev_is_pci(parent))
2287                 return NULL; /* not a PCI device */
2288
2289         pdev = to_pci_dev(parent);
2290         if (!pdev->slot) {
2291                 netdev_notice(vf_netdev, "no PCI slot information\n");
2292                 return NULL;
2293         }
2294
2295         if (kstrtou32(pci_slot_name(pdev->slot), 10, &serial)) {
2296                 netdev_notice(vf_netdev, "Invalid vf serial:%s\n",
2297                               pci_slot_name(pdev->slot));
2298                 return NULL;
2299         }
2300
2301         list_for_each_entry(ndev_ctx, &netvsc_dev_list, list) {
2302                 if (!ndev_ctx->vf_alloc)
2303                         continue;
2304
2305                 if (ndev_ctx->vf_serial == serial)
2306                         return hv_get_drvdata(ndev_ctx->device_ctx);
2307         }
2308
2309         netdev_notice(vf_netdev,
2310                       "no netdev found for vf serial:%u\n", serial);
2311         return NULL;
2312 }
2313
2314 static int netvsc_register_vf(struct net_device *vf_netdev)
2315 {
2316         struct net_device_context *net_device_ctx;
2317         struct netvsc_device *netvsc_dev;
2318         struct bpf_prog *prog;
2319         struct net_device *ndev;
2320         int ret;
2321
2322         if (vf_netdev->addr_len != ETH_ALEN)
2323                 return NOTIFY_DONE;
2324
2325         ndev = get_netvsc_byslot(vf_netdev);
2326         if (!ndev)
2327                 return NOTIFY_DONE;
2328
2329         net_device_ctx = netdev_priv(ndev);
2330         netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2331         if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
2332                 return NOTIFY_DONE;
2333
2334         /* if synthetic interface is a different namespace,
2335          * then move the VF to that namespace; join will be
2336          * done again in that context.
2337          */
2338         if (!net_eq(dev_net(ndev), dev_net(vf_netdev))) {
2339                 ret = dev_change_net_namespace(vf_netdev,
2340                                                dev_net(ndev), "eth%d");
2341                 if (ret)
2342                         netdev_err(vf_netdev,
2343                                    "could not move to same namespace as %s: %d\n",
2344                                    ndev->name, ret);
2345                 else
2346                         netdev_info(vf_netdev,
2347                                     "VF moved to namespace with: %s\n",
2348                                     ndev->name);
2349                 return NOTIFY_DONE;
2350         }
2351
2352         netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
2353
2354         if (netvsc_vf_join(vf_netdev, ndev) != 0)
2355                 return NOTIFY_DONE;
2356
2357         dev_hold(vf_netdev);
2358         rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
2359
2360         vf_netdev->wanted_features = ndev->features;
2361         netdev_update_features(vf_netdev);
2362
2363         prog = netvsc_xdp_get(netvsc_dev);
2364         netvsc_vf_setxdp(vf_netdev, prog);
2365
2366         return NOTIFY_OK;
2367 }
2368
2369 /* VF up/down change detected, schedule to change data path */
2370 static int netvsc_vf_changed(struct net_device *vf_netdev)
2371 {
2372         struct net_device_context *net_device_ctx;
2373         struct netvsc_device *netvsc_dev;
2374         struct net_device *ndev;
2375         bool vf_is_up = netif_running(vf_netdev);
2376
2377         ndev = get_netvsc_byref(vf_netdev);
2378         if (!ndev)
2379                 return NOTIFY_DONE;
2380
2381         net_device_ctx = netdev_priv(ndev);
2382         netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2383         if (!netvsc_dev)
2384                 return NOTIFY_DONE;
2385
2386         netvsc_switch_datapath(ndev, vf_is_up);
2387         netdev_info(ndev, "Data path switched %s VF: %s\n",
2388                     vf_is_up ? "to" : "from", vf_netdev->name);
2389
2390         return NOTIFY_OK;
2391 }
2392
2393 static int netvsc_unregister_vf(struct net_device *vf_netdev)
2394 {
2395         struct net_device *ndev;
2396         struct net_device_context *net_device_ctx;
2397
2398         ndev = get_netvsc_byref(vf_netdev);
2399         if (!ndev)
2400                 return NOTIFY_DONE;
2401
2402         net_device_ctx = netdev_priv(ndev);
2403         cancel_delayed_work_sync(&net_device_ctx->vf_takeover);
2404
2405         netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
2406
2407         netvsc_vf_setxdp(vf_netdev, NULL);
2408
2409         netdev_rx_handler_unregister(vf_netdev);
2410         netdev_upper_dev_unlink(vf_netdev, ndev);
2411         RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
2412         dev_put(vf_netdev);
2413
2414         return NOTIFY_OK;
2415 }
2416
2417 static int netvsc_probe(struct hv_device *dev,
2418                         const struct hv_vmbus_device_id *dev_id)
2419 {
2420         struct net_device *net = NULL;
2421         struct net_device_context *net_device_ctx;
2422         struct netvsc_device_info *device_info = NULL;
2423         struct netvsc_device *nvdev;
2424         int ret = -ENOMEM;
2425
2426         net = alloc_etherdev_mq(sizeof(struct net_device_context),
2427                                 VRSS_CHANNEL_MAX);
2428         if (!net)
2429                 goto no_net;
2430
2431         netif_carrier_off(net);
2432
2433         netvsc_init_settings(net);
2434
2435         net_device_ctx = netdev_priv(net);
2436         net_device_ctx->device_ctx = dev;
2437         net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
2438         if (netif_msg_probe(net_device_ctx))
2439                 netdev_dbg(net, "netvsc msg_enable: %d\n",
2440                            net_device_ctx->msg_enable);
2441
2442         hv_set_drvdata(dev, net);
2443
2444         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
2445
2446         spin_lock_init(&net_device_ctx->lock);
2447         INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
2448         INIT_DELAYED_WORK(&net_device_ctx->vf_takeover, netvsc_vf_setup);
2449
2450         net_device_ctx->vf_stats
2451                 = netdev_alloc_pcpu_stats(struct netvsc_vf_pcpu_stats);
2452         if (!net_device_ctx->vf_stats)
2453                 goto no_stats;
2454
2455         net->netdev_ops = &device_ops;
2456         net->ethtool_ops = &ethtool_ops;
2457         SET_NETDEV_DEV(net, &dev->device);
2458
2459         /* We always need headroom for rndis header */
2460         net->needed_headroom = RNDIS_AND_PPI_SIZE;
2461
2462         /* Initialize the number of queues to be 1, we may change it if more
2463          * channels are offered later.
2464          */
2465         netif_set_real_num_tx_queues(net, 1);
2466         netif_set_real_num_rx_queues(net, 1);
2467
2468         /* Notify the netvsc driver of the new device */
2469         device_info = netvsc_devinfo_get(NULL);
2470
2471         if (!device_info) {
2472                 ret = -ENOMEM;
2473                 goto devinfo_failed;
2474         }
2475
2476         nvdev = rndis_filter_device_add(dev, device_info);
2477         if (IS_ERR(nvdev)) {
2478                 ret = PTR_ERR(nvdev);
2479                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
2480                 goto rndis_failed;
2481         }
2482
2483         memcpy(net->dev_addr, device_info->mac_adr, ETH_ALEN);
2484
2485         /* We must get rtnl lock before scheduling nvdev->subchan_work,
2486          * otherwise netvsc_subchan_work() can get rtnl lock first and wait
2487          * all subchannels to show up, but that may not happen because
2488          * netvsc_probe() can't get rtnl lock and as a result vmbus_onoffer()
2489          * -> ... -> device_add() -> ... -> __device_attach() can't get
2490          * the device lock, so all the subchannels can't be processed --
2491          * finally netvsc_subchan_work() hangs forever.
2492          */
2493         rtnl_lock();
2494
2495         if (nvdev->num_chn > 1)
2496                 schedule_work(&nvdev->subchan_work);
2497
2498         /* hw_features computed in rndis_netdev_set_hwcaps() */
2499         net->features = net->hw_features |
2500                 NETIF_F_HIGHDMA | NETIF_F_HW_VLAN_CTAG_TX |
2501                 NETIF_F_HW_VLAN_CTAG_RX;
2502         net->vlan_features = net->features;
2503
2504         netdev_lockdep_set_classes(net);
2505
2506         /* MTU range: 68 - 1500 or 65521 */
2507         net->min_mtu = NETVSC_MTU_MIN;
2508         if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
2509                 net->max_mtu = NETVSC_MTU - ETH_HLEN;
2510         else
2511                 net->max_mtu = ETH_DATA_LEN;
2512
2513         nvdev->tx_disable = false;
2514
2515         ret = register_netdevice(net);
2516         if (ret != 0) {
2517                 pr_err("Unable to register netdev.\n");
2518                 goto register_failed;
2519         }
2520
2521         list_add(&net_device_ctx->list, &netvsc_dev_list);
2522         rtnl_unlock();
2523
2524         netvsc_devinfo_put(device_info);
2525         return 0;
2526
2527 register_failed:
2528         rtnl_unlock();
2529         rndis_filter_device_remove(dev, nvdev);
2530 rndis_failed:
2531         netvsc_devinfo_put(device_info);
2532 devinfo_failed:
2533         free_percpu(net_device_ctx->vf_stats);
2534 no_stats:
2535         hv_set_drvdata(dev, NULL);
2536         free_netdev(net);
2537 no_net:
2538         return ret;
2539 }
2540
2541 static int netvsc_remove(struct hv_device *dev)
2542 {
2543         struct net_device_context *ndev_ctx;
2544         struct net_device *vf_netdev, *net;
2545         struct netvsc_device *nvdev;
2546
2547         net = hv_get_drvdata(dev);
2548         if (net == NULL) {
2549                 dev_err(&dev->device, "No net device to remove\n");
2550                 return 0;
2551         }
2552
2553         ndev_ctx = netdev_priv(net);
2554
2555         cancel_delayed_work_sync(&ndev_ctx->dwork);
2556
2557         rtnl_lock();
2558         nvdev = rtnl_dereference(ndev_ctx->nvdev);
2559         if (nvdev) {
2560                 cancel_work_sync(&nvdev->subchan_work);
2561                 netvsc_xdp_set(net, NULL, NULL, nvdev);
2562         }
2563
2564         /*
2565          * Call to the vsc driver to let it know that the device is being
2566          * removed. Also blocks mtu and channel changes.
2567          */
2568         vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2569         if (vf_netdev)
2570                 netvsc_unregister_vf(vf_netdev);
2571
2572         if (nvdev)
2573                 rndis_filter_device_remove(dev, nvdev);
2574
2575         unregister_netdevice(net);
2576         list_del(&ndev_ctx->list);
2577
2578         rtnl_unlock();
2579
2580         hv_set_drvdata(dev, NULL);
2581
2582         free_percpu(ndev_ctx->vf_stats);
2583         free_netdev(net);
2584         return 0;
2585 }
2586
2587 static int netvsc_suspend(struct hv_device *dev)
2588 {
2589         struct net_device_context *ndev_ctx;
2590         struct net_device *vf_netdev, *net;
2591         struct netvsc_device *nvdev;
2592         int ret;
2593
2594         net = hv_get_drvdata(dev);
2595
2596         ndev_ctx = netdev_priv(net);
2597         cancel_delayed_work_sync(&ndev_ctx->dwork);
2598
2599         rtnl_lock();
2600
2601         nvdev = rtnl_dereference(ndev_ctx->nvdev);
2602         if (nvdev == NULL) {
2603                 ret = -ENODEV;
2604                 goto out;
2605         }
2606
2607         vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2608         if (vf_netdev)
2609                 netvsc_unregister_vf(vf_netdev);
2610
2611         /* Save the current config info */
2612         ndev_ctx->saved_netvsc_dev_info = netvsc_devinfo_get(nvdev);
2613
2614         ret = netvsc_detach(net, nvdev);
2615 out:
2616         rtnl_unlock();
2617
2618         return ret;
2619 }
2620
2621 static int netvsc_resume(struct hv_device *dev)
2622 {
2623         struct net_device *net = hv_get_drvdata(dev);
2624         struct net_device_context *net_device_ctx;
2625         struct netvsc_device_info *device_info;
2626         int ret;
2627
2628         rtnl_lock();
2629
2630         net_device_ctx = netdev_priv(net);
2631         device_info = net_device_ctx->saved_netvsc_dev_info;
2632
2633         ret = netvsc_attach(net, device_info);
2634
2635         netvsc_devinfo_put(device_info);
2636         net_device_ctx->saved_netvsc_dev_info = NULL;
2637
2638         rtnl_unlock();
2639
2640         return ret;
2641 }
2642 static const struct hv_vmbus_device_id id_table[] = {
2643         /* Network guid */
2644         { HV_NIC_GUID, },
2645         { },
2646 };
2647
2648 MODULE_DEVICE_TABLE(vmbus, id_table);
2649
2650 /* The one and only one */
2651 static struct  hv_driver netvsc_drv = {
2652         .name = KBUILD_MODNAME,
2653         .id_table = id_table,
2654         .probe = netvsc_probe,
2655         .remove = netvsc_remove,
2656         .suspend = netvsc_suspend,
2657         .resume = netvsc_resume,
2658         .driver = {
2659                 .probe_type = PROBE_FORCE_SYNCHRONOUS,
2660         },
2661 };
2662
2663 /*
2664  * On Hyper-V, every VF interface is matched with a corresponding
2665  * synthetic interface. The synthetic interface is presented first
2666  * to the guest. When the corresponding VF instance is registered,
2667  * we will take care of switching the data path.
2668  */
2669 static int netvsc_netdev_event(struct notifier_block *this,
2670                                unsigned long event, void *ptr)
2671 {
2672         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2673
2674         /* Skip our own events */
2675         if (event_dev->netdev_ops == &device_ops)
2676                 return NOTIFY_DONE;
2677
2678         /* Avoid non-Ethernet type devices */
2679         if (event_dev->type != ARPHRD_ETHER)
2680                 return NOTIFY_DONE;
2681
2682         /* Avoid Vlan dev with same MAC registering as VF */
2683         if (is_vlan_dev(event_dev))
2684                 return NOTIFY_DONE;
2685
2686         /* Avoid Bonding master dev with same MAC registering as VF */
2687         if ((event_dev->priv_flags & IFF_BONDING) &&
2688             (event_dev->flags & IFF_MASTER))
2689                 return NOTIFY_DONE;
2690
2691         switch (event) {
2692         case NETDEV_REGISTER:
2693                 return netvsc_register_vf(event_dev);
2694         case NETDEV_UNREGISTER:
2695                 return netvsc_unregister_vf(event_dev);
2696         case NETDEV_UP:
2697         case NETDEV_DOWN:
2698                 return netvsc_vf_changed(event_dev);
2699         default:
2700                 return NOTIFY_DONE;
2701         }
2702 }
2703
2704 static struct notifier_block netvsc_netdev_notifier = {
2705         .notifier_call = netvsc_netdev_event,
2706 };
2707
2708 static void __exit netvsc_drv_exit(void)
2709 {
2710         unregister_netdevice_notifier(&netvsc_netdev_notifier);
2711         vmbus_driver_unregister(&netvsc_drv);
2712 }
2713
2714 static int __init netvsc_drv_init(void)
2715 {
2716         int ret;
2717
2718         if (ring_size < RING_SIZE_MIN) {
2719                 ring_size = RING_SIZE_MIN;
2720                 pr_info("Increased ring_size to %u (min allowed)\n",
2721                         ring_size);
2722         }
2723         netvsc_ring_bytes = ring_size * PAGE_SIZE;
2724
2725         ret = vmbus_driver_register(&netvsc_drv);
2726         if (ret)
2727                 return ret;
2728
2729         register_netdevice_notifier(&netvsc_netdev_notifier);
2730         return 0;
2731 }
2732
2733 MODULE_LICENSE("GPL");
2734 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
2735
2736 module_init(netvsc_drv_init);
2737 module_exit(netvsc_drv_exit);