drivers/net/hyperv/netvsc_drv.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (c) 2009, Microsoft Corporation.
   4  *
   5  * Authors:
   6  *   Haiyang Zhang <haiyangz@microsoft.com>
   7  *   Hank Janssen  <hjanssen@microsoft.com>
   8  */
   9 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  10
  11 #include <linux/init.h>
  12 #include <linux/atomic.h>
  13 #include <linux/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 (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(unsigned long hvpfn, u32 offset, u32 len,
 377                        struct hv_page_buffer *pb)
 378 {
 379         int j = 0;
 380
 381         hvpfn += offset >> HV_HYP_PAGE_SHIFT;
 382         offset = offset & ~HV_HYP_PAGE_MASK;
 383
 384         while (len > 0) {
 385                 unsigned long bytes;
 386
 387                 bytes = HV_HYP_PAGE_SIZE - offset;
 388                 if (bytes > len)
 389                         bytes = len;
 390                 pb[j].pfn = hvpfn;
 391                 pb[j].offset = offset;
 392                 pb[j].len = bytes;
 393
 394                 offset += bytes;
 395                 len -= bytes;
 396
 397                 if (offset == HV_HYP_PAGE_SIZE && len) {
 398                         hvpfn++;
 399                         offset = 0;
 400                         j++;
 401                 }
 402         }
 403
 404         return j + 1;
 405 }
 406
 407 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
 408                            struct hv_netvsc_packet *packet,
 409                            struct hv_page_buffer *pb)
 410 {
 411         u32 slots_used = 0;
 412         char *data = skb->data;
 413         int frags = skb_shinfo(skb)->nr_frags;
 414         int i;
 415
 416         /* The packet is laid out thus:
 417          * 1. hdr: RNDIS header and PPI
 418          * 2. skb linear data
 419          * 3. skb fragment data
 420          */
 421         slots_used += fill_pg_buf(virt_to_hvpfn(hdr),
 422                                   offset_in_hvpage(hdr),
 423                                   len,
 424                                   &pb[slots_used]);
 425
 426         packet->rmsg_size = len;
 427         packet->rmsg_pgcnt = slots_used;
 428
 429         slots_used += fill_pg_buf(virt_to_hvpfn(data),
 430                                   offset_in_hvpage(data),
 431                                   skb_headlen(skb),
 432                                   &pb[slots_used]);
 433
 434         for (i = 0; i < frags; i++) {
 435                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 436
 437                 slots_used += fill_pg_buf(page_to_hvpfn(skb_frag_page(frag)),
 438                                           skb_frag_off(frag),
 439                                           skb_frag_size(frag),
 440                                           &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 &= ~HV_HYP_PAGE_MASK;
 457                 pages += HVPFN_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_hvpage(data);
 466         unsigned int len = skb_headlen(skb);
 467         int slots;
 468         int frag_slots;
 469
 470         slots = DIV_ROUND_UP(offset + len, HV_HYP_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_record_rx_queue(skb, 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         /* Ensure the packet is big enough to access its fields */
 752         if (resp->msg_len - RNDIS_HEADER_SIZE < sizeof(struct rndis_indicate_status)) {
 753                 netdev_err(net, "invalid rndis_indicate_status packet, len: %u\n",
 754                            resp->msg_len);
 755                 return;
 756         }
 757
 758         /* Update the physical link speed when changing to another vSwitch */
 759         if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
 760                 u32 speed;
 761
 762                 speed = *(u32 *)((void *)indicate
 763                                  + indicate->status_buf_offset) / 10000;
 764                 ndev_ctx->speed = speed;
 765                 return;
 766         }
 767
 768         /* Handle these link change statuses below */
 769         if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
 770             indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
 771             indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
 772                 return;
 773
 774         if (net->reg_state != NETREG_REGISTERED)
 775                 return;
 776
 777         event = kzalloc(sizeof(*event), GFP_ATOMIC);
 778         if (!event)
 779                 return;
 780         event->event = indicate->status;
 781
 782         spin_lock_irqsave(&ndev_ctx->lock, flags);
 783         list_add_tail(&event->list, &ndev_ctx->reconfig_events);
 784         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
 785
 786         schedule_delayed_work(&ndev_ctx->dwork, 0);
 787 }
 788
 789 static void netvsc_xdp_xmit(struct sk_buff *skb, struct net_device *ndev)
 790 {
 791         int rc;
 792
 793         skb->queue_mapping = skb_get_rx_queue(skb);
 794         __skb_push(skb, ETH_HLEN);
 795
 796         rc = netvsc_xmit(skb, ndev, true);
 797
 798         if (dev_xmit_complete(rc))
 799                 return;
 800
 801         dev_kfree_skb_any(skb);
 802         ndev->stats.tx_dropped++;
 803 }
 804
 805 static void netvsc_comp_ipcsum(struct sk_buff *skb)
 806 {
 807         struct iphdr *iph = (struct iphdr *)skb->data;
 808
 809         iph->check = 0;
 810         iph->check = ip_fast_csum(iph, iph->ihl);
 811 }
 812
 813 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
 814                                              struct netvsc_channel *nvchan,
 815                                              struct xdp_buff *xdp)
 816 {
 817         struct napi_struct *napi = &nvchan->napi;
 818         const struct ndis_pkt_8021q_info *vlan = nvchan->rsc.vlan;
 819         const struct ndis_tcp_ip_checksum_info *csum_info =
 820                                                 nvchan->rsc.csum_info;
 821         const u32 *hash_info = nvchan->rsc.hash_info;
 822         struct sk_buff *skb;
 823         void *xbuf = xdp->data_hard_start;
 824         int i;
 825
 826         if (xbuf) {
 827                 unsigned int hdroom = xdp->data - xdp->data_hard_start;
 828                 unsigned int xlen = xdp->data_end - xdp->data;
 829                 unsigned int frag_size = xdp->frame_sz;
 830
 831                 skb = build_skb(xbuf, frag_size);
 832
 833                 if (!skb) {
 834                         __free_page(virt_to_page(xbuf));
 835                         return NULL;
 836                 }
 837
 838                 skb_reserve(skb, hdroom);
 839                 skb_put(skb, xlen);
 840                 skb->dev = napi->dev;
 841         } else {
 842                 skb = napi_alloc_skb(napi, nvchan->rsc.pktlen);
 843
 844                 if (!skb)
 845                         return NULL;
 846
 847                 /* Copy to skb. This copy is needed here since the memory
 848                  * pointed by hv_netvsc_packet cannot be deallocated.
 849                  */
 850                 for (i = 0; i < nvchan->rsc.cnt; i++)
 851                         skb_put_data(skb, nvchan->rsc.data[i],
 852                                      nvchan->rsc.len[i]);
 853         }
 854
 855         skb->protocol = eth_type_trans(skb, net);
 856
 857         /* skb is already created with CHECKSUM_NONE */
 858         skb_checksum_none_assert(skb);
 859
 860         /* Incoming packets may have IP header checksum verified by the host.
 861          * They may not have IP header checksum computed after coalescing.
 862          * We compute it here if the flags are set, because on Linux, the IP
 863          * checksum is always checked.
 864          */
 865         if (csum_info && csum_info->receive.ip_checksum_value_invalid &&
 866             csum_info->receive.ip_checksum_succeeded &&
 867             skb->protocol == htons(ETH_P_IP))
 868                 netvsc_comp_ipcsum(skb);
 869
 870         /* Do L4 checksum offload if enabled and present. */
 871         if (csum_info && (net->features & NETIF_F_RXCSUM)) {
 872                 if (csum_info->receive.tcp_checksum_succeeded ||
 873                     csum_info->receive.udp_checksum_succeeded)
 874                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 875         }
 876
 877         if (hash_info && (net->features & NETIF_F_RXHASH))
 878                 skb_set_hash(skb, *hash_info, PKT_HASH_TYPE_L4);
 879
 880         if (vlan) {
 881                 u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT) |
 882                         (vlan->cfi ? VLAN_CFI_MASK : 0);
 883
 884                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 885                                        vlan_tci);
 886         }
 887
 888         return skb;
 889 }
 890
 891 /*
 892  * netvsc_recv_callback -  Callback when we receive a packet from the
 893  * "wire" on the specified device.
 894  */
 895 int netvsc_recv_callback(struct net_device *net,
 896                          struct netvsc_device *net_device,
 897                          struct netvsc_channel *nvchan)
 898 {
 899         struct net_device_context *net_device_ctx = netdev_priv(net);
 900         struct vmbus_channel *channel = nvchan->channel;
 901         u16 q_idx = channel->offermsg.offer.sub_channel_index;
 902         struct sk_buff *skb;
 903         struct netvsc_stats *rx_stats = &nvchan->rx_stats;
 904         struct xdp_buff xdp;
 905         u32 act;
 906
 907         if (net->reg_state != NETREG_REGISTERED)
 908                 return NVSP_STAT_FAIL;
 909
 910         act = netvsc_run_xdp(net, nvchan, &xdp);
 911
 912         if (act != XDP_PASS && act != XDP_TX) {
 913                 u64_stats_update_begin(&rx_stats->syncp);
 914                 rx_stats->xdp_drop++;
 915                 u64_stats_update_end(&rx_stats->syncp);
 916
 917                 return NVSP_STAT_SUCCESS; /* consumed by XDP */
 918         }
 919
 920         /* Allocate a skb - TODO direct I/O to pages? */
 921         skb = netvsc_alloc_recv_skb(net, nvchan, &xdp);
 922
 923         if (unlikely(!skb)) {
 924                 ++net_device_ctx->eth_stats.rx_no_memory;
 925                 return NVSP_STAT_FAIL;
 926         }
 927
 928         skb_record_rx_queue(skb, q_idx);
 929
 930         /*
 931          * Even if injecting the packet, record the statistics
 932          * on the synthetic device because modifying the VF device
 933          * statistics will not work correctly.
 934          */
 935         u64_stats_update_begin(&rx_stats->syncp);
 936         rx_stats->packets++;
 937         rx_stats->bytes += nvchan->rsc.pktlen;
 938
 939         if (skb->pkt_type == PACKET_BROADCAST)
 940                 ++rx_stats->broadcast;
 941         else if (skb->pkt_type == PACKET_MULTICAST)
 942                 ++rx_stats->multicast;
 943         u64_stats_update_end(&rx_stats->syncp);
 944
 945         if (act == XDP_TX) {
 946                 netvsc_xdp_xmit(skb, net);
 947                 return NVSP_STAT_SUCCESS;
 948         }
 949
 950         napi_gro_receive(&nvchan->napi, skb);
 951         return NVSP_STAT_SUCCESS;
 952 }
 953
 954 static void netvsc_get_drvinfo(struct net_device *net,
 955                                struct ethtool_drvinfo *info)
 956 {
 957         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
 958         strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
 959 }
 960
 961 static void netvsc_get_channels(struct net_device *net,
 962                                 struct ethtool_channels *channel)
 963 {
 964         struct net_device_context *net_device_ctx = netdev_priv(net);
 965         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
 966
 967         if (nvdev) {
 968                 channel->max_combined   = nvdev->max_chn;
 969                 channel->combined_count = nvdev->num_chn;
 970         }
 971 }
 972
 973 /* Alloc struct netvsc_device_info, and initialize it from either existing
 974  * struct netvsc_device, or from default values.
 975  */
 976 static
 977 struct netvsc_device_info *netvsc_devinfo_get(struct netvsc_device *nvdev)
 978 {
 979         struct netvsc_device_info *dev_info;
 980         struct bpf_prog *prog;
 981
 982         dev_info = kzalloc(sizeof(*dev_info), GFP_ATOMIC);
 983
 984         if (!dev_info)
 985                 return NULL;
 986
 987         if (nvdev) {
 988                 ASSERT_RTNL();
 989
 990                 dev_info->num_chn = nvdev->num_chn;
 991                 dev_info->send_sections = nvdev->send_section_cnt;
 992                 dev_info->send_section_size = nvdev->send_section_size;
 993                 dev_info->recv_sections = nvdev->recv_section_cnt;
 994                 dev_info->recv_section_size = nvdev->recv_section_size;
 995
 996                 memcpy(dev_info->rss_key, nvdev->extension->rss_key,
 997                        NETVSC_HASH_KEYLEN);
 998
 999                 prog = netvsc_xdp_get(nvdev);
1000                 if (prog) {
1001                         bpf_prog_inc(prog);
1002                         dev_info->bprog = prog;
1003                 }
1004         } else {
1005                 dev_info->num_chn = VRSS_CHANNEL_DEFAULT;
1006                 dev_info->send_sections = NETVSC_DEFAULT_TX;
1007                 dev_info->send_section_size = NETVSC_SEND_SECTION_SIZE;
1008                 dev_info->recv_sections = NETVSC_DEFAULT_RX;
1009                 dev_info->recv_section_size = NETVSC_RECV_SECTION_SIZE;
1010         }
1011
1012         return dev_info;
1013 }
1014
1015 /* Free struct netvsc_device_info */
1016 static void netvsc_devinfo_put(struct netvsc_device_info *dev_info)
1017 {
1018         if (dev_info->bprog) {
1019                 ASSERT_RTNL();
1020                 bpf_prog_put(dev_info->bprog);
1021         }
1022
1023         kfree(dev_info);
1024 }
1025
1026 static int netvsc_detach(struct net_device *ndev,
1027                          struct netvsc_device *nvdev)
1028 {
1029         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1030         struct hv_device *hdev = ndev_ctx->device_ctx;
1031         int ret;
1032
1033         /* Don't try continuing to try and setup sub channels */
1034         if (cancel_work_sync(&nvdev->subchan_work))
1035                 nvdev->num_chn = 1;
1036
1037         netvsc_xdp_set(ndev, NULL, NULL, nvdev);
1038
1039         /* If device was up (receiving) then shutdown */
1040         if (netif_running(ndev)) {
1041                 netvsc_tx_disable(nvdev, ndev);
1042
1043                 ret = rndis_filter_close(nvdev);
1044                 if (ret) {
1045                         netdev_err(ndev,
1046                                    "unable to close device (ret %d).\n", ret);
1047                         return ret;
1048                 }
1049
1050                 ret = netvsc_wait_until_empty(nvdev);
1051                 if (ret) {
1052                         netdev_err(ndev,
1053                                    "Ring buffer not empty after closing rndis\n");
1054                         return ret;
1055                 }
1056         }
1057
1058         netif_device_detach(ndev);
1059
1060         rndis_filter_device_remove(hdev, nvdev);
1061
1062         return 0;
1063 }
1064
1065 static int netvsc_attach(struct net_device *ndev,
1066                          struct netvsc_device_info *dev_info)
1067 {
1068         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1069         struct hv_device *hdev = ndev_ctx->device_ctx;
1070         struct netvsc_device *nvdev;
1071         struct rndis_device *rdev;
1072         struct bpf_prog *prog;
1073         int ret = 0;
1074
1075         nvdev = rndis_filter_device_add(hdev, dev_info);
1076         if (IS_ERR(nvdev))
1077                 return PTR_ERR(nvdev);
1078
1079         if (nvdev->num_chn > 1) {
1080                 ret = rndis_set_subchannel(ndev, nvdev, dev_info);
1081
1082                 /* if unavailable, just proceed with one queue */
1083                 if (ret) {
1084                         nvdev->max_chn = 1;
1085                         nvdev->num_chn = 1;
1086                 }
1087         }
1088
1089         prog = dev_info->bprog;
1090         if (prog) {
1091                 bpf_prog_inc(prog);
1092                 ret = netvsc_xdp_set(ndev, prog, NULL, nvdev);
1093                 if (ret) {
1094                         bpf_prog_put(prog);
1095                         goto err1;
1096                 }
1097         }
1098
1099         /* In any case device is now ready */
1100         nvdev->tx_disable = false;
1101         netif_device_attach(ndev);
1102
1103         /* Note: enable and attach happen when sub-channels setup */
1104         netif_carrier_off(ndev);
1105
1106         if (netif_running(ndev)) {
1107                 ret = rndis_filter_open(nvdev);
1108                 if (ret)
1109                         goto err2;
1110
1111                 rdev = nvdev->extension;
1112                 if (!rdev->link_state)
1113                         netif_carrier_on(ndev);
1114         }
1115
1116         return 0;
1117
1118 err2:
1119         netif_device_detach(ndev);
1120
1121 err1:
1122         rndis_filter_device_remove(hdev, nvdev);
1123
1124         return ret;
1125 }
1126
1127 static int netvsc_set_channels(struct net_device *net,
1128                                struct ethtool_channels *channels)
1129 {
1130         struct net_device_context *net_device_ctx = netdev_priv(net);
1131         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
1132         unsigned int orig, count = channels->combined_count;
1133         struct netvsc_device_info *device_info;
1134         int ret;
1135
1136         /* We do not support separate count for rx, tx, or other */
1137         if (count == 0 ||
1138             channels->rx_count || channels->tx_count || channels->other_count)
1139                 return -EINVAL;
1140
1141         if (!nvdev || nvdev->destroy)
1142                 return -ENODEV;
1143
1144         if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
1145                 return -EINVAL;
1146
1147         if (count > nvdev->max_chn)
1148                 return -EINVAL;
1149
1150         orig = nvdev->num_chn;
1151
1152         device_info = netvsc_devinfo_get(nvdev);
1153
1154         if (!device_info)
1155                 return -ENOMEM;
1156
1157         device_info->num_chn = count;
1158
1159         ret = netvsc_detach(net, nvdev);
1160         if (ret)
1161                 goto out;
1162
1163         ret = netvsc_attach(net, device_info);
1164         if (ret) {
1165                 device_info->num_chn = orig;
1166                 if (netvsc_attach(net, device_info))
1167                         netdev_err(net, "restoring channel setting failed\n");
1168         }
1169
1170 out:
1171         netvsc_devinfo_put(device_info);
1172         return ret;
1173 }
1174
1175 static void netvsc_init_settings(struct net_device *dev)
1176 {
1177         struct net_device_context *ndc = netdev_priv(dev);
1178
1179         ndc->l4_hash = HV_DEFAULT_L4HASH;
1180
1181         ndc->speed = SPEED_UNKNOWN;
1182         ndc->duplex = DUPLEX_FULL;
1183
1184         dev->features = NETIF_F_LRO;
1185 }
1186
1187 static int netvsc_get_link_ksettings(struct net_device *dev,
1188                                      struct ethtool_link_ksettings *cmd)
1189 {
1190         struct net_device_context *ndc = netdev_priv(dev);
1191         struct net_device *vf_netdev;
1192
1193         vf_netdev = rtnl_dereference(ndc->vf_netdev);
1194
1195         if (vf_netdev)
1196                 return __ethtool_get_link_ksettings(vf_netdev, cmd);
1197
1198         cmd->base.speed = ndc->speed;
1199         cmd->base.duplex = ndc->duplex;
1200         cmd->base.port = PORT_OTHER;
1201
1202         return 0;
1203 }
1204
1205 static int netvsc_set_link_ksettings(struct net_device *dev,
1206                                      const struct ethtool_link_ksettings *cmd)
1207 {
1208         struct net_device_context *ndc = netdev_priv(dev);
1209         struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1210
1211         if (vf_netdev) {
1212                 if (!vf_netdev->ethtool_ops->set_link_ksettings)
1213                         return -EOPNOTSUPP;
1214
1215                 return vf_netdev->ethtool_ops->set_link_ksettings(vf_netdev,
1216                                                                   cmd);
1217         }
1218
1219         return ethtool_virtdev_set_link_ksettings(dev, cmd,
1220                                                   &ndc->speed, &ndc->duplex);
1221 }
1222
1223 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
1224 {
1225         struct net_device_context *ndevctx = netdev_priv(ndev);
1226         struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1227         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1228         int orig_mtu = ndev->mtu;
1229         struct netvsc_device_info *device_info;
1230         int ret = 0;
1231
1232         if (!nvdev || nvdev->destroy)
1233                 return -ENODEV;
1234
1235         device_info = netvsc_devinfo_get(nvdev);
1236
1237         if (!device_info)
1238                 return -ENOMEM;
1239
1240         /* Change MTU of underlying VF netdev first. */
1241         if (vf_netdev) {
1242                 ret = dev_set_mtu(vf_netdev, mtu);
1243                 if (ret)
1244                         goto out;
1245         }
1246
1247         ret = netvsc_detach(ndev, nvdev);
1248         if (ret)
1249                 goto rollback_vf;
1250
1251         ndev->mtu = mtu;
1252
1253         ret = netvsc_attach(ndev, device_info);
1254         if (!ret)
1255                 goto out;
1256
1257         /* Attempt rollback to original MTU */
1258         ndev->mtu = orig_mtu;
1259
1260         if (netvsc_attach(ndev, device_info))
1261                 netdev_err(ndev, "restoring mtu failed\n");
1262 rollback_vf:
1263         if (vf_netdev)
1264                 dev_set_mtu(vf_netdev, orig_mtu);
1265
1266 out:
1267         netvsc_devinfo_put(device_info);
1268         return ret;
1269 }
1270
1271 static void netvsc_get_vf_stats(struct net_device *net,
1272                                 struct netvsc_vf_pcpu_stats *tot)
1273 {
1274         struct net_device_context *ndev_ctx = netdev_priv(net);
1275         int i;
1276
1277         memset(tot, 0, sizeof(*tot));
1278
1279         for_each_possible_cpu(i) {
1280                 const struct netvsc_vf_pcpu_stats *stats
1281                         = per_cpu_ptr(ndev_ctx->vf_stats, i);
1282                 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1283                 unsigned int start;
1284
1285                 do {
1286                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1287                         rx_packets = stats->rx_packets;
1288                         tx_packets = stats->tx_packets;
1289                         rx_bytes = stats->rx_bytes;
1290                         tx_bytes = stats->tx_bytes;
1291                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1292
1293                 tot->rx_packets += rx_packets;
1294                 tot->tx_packets += tx_packets;
1295                 tot->rx_bytes   += rx_bytes;
1296                 tot->tx_bytes   += tx_bytes;
1297                 tot->tx_dropped += stats->tx_dropped;
1298         }
1299 }
1300
1301 static void netvsc_get_pcpu_stats(struct net_device *net,
1302                                   struct netvsc_ethtool_pcpu_stats *pcpu_tot)
1303 {
1304         struct net_device_context *ndev_ctx = netdev_priv(net);
1305         struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
1306         int i;
1307
1308         /* fetch percpu stats of vf */
1309         for_each_possible_cpu(i) {
1310                 const struct netvsc_vf_pcpu_stats *stats =
1311                         per_cpu_ptr(ndev_ctx->vf_stats, i);
1312                 struct netvsc_ethtool_pcpu_stats *this_tot = &pcpu_tot[i];
1313                 unsigned int start;
1314
1315                 do {
1316                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1317                         this_tot->vf_rx_packets = stats->rx_packets;
1318                         this_tot->vf_tx_packets = stats->tx_packets;
1319                         this_tot->vf_rx_bytes = stats->rx_bytes;
1320                         this_tot->vf_tx_bytes = stats->tx_bytes;
1321                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1322                 this_tot->rx_packets = this_tot->vf_rx_packets;
1323                 this_tot->tx_packets = this_tot->vf_tx_packets;
1324                 this_tot->rx_bytes   = this_tot->vf_rx_bytes;
1325                 this_tot->tx_bytes   = this_tot->vf_tx_bytes;
1326         }
1327
1328         /* fetch percpu stats of netvsc */
1329         for (i = 0; i < nvdev->num_chn; i++) {
1330                 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1331                 const struct netvsc_stats *stats;
1332                 struct netvsc_ethtool_pcpu_stats *this_tot =
1333                         &pcpu_tot[nvchan->channel->target_cpu];
1334                 u64 packets, bytes;
1335                 unsigned int start;
1336
1337                 stats = &nvchan->tx_stats;
1338                 do {
1339                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1340                         packets = stats->packets;
1341                         bytes = stats->bytes;
1342                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1343
1344                 this_tot->tx_bytes      += bytes;
1345                 this_tot->tx_packets    += packets;
1346
1347                 stats = &nvchan->rx_stats;
1348                 do {
1349                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1350                         packets = stats->packets;
1351                         bytes = stats->bytes;
1352                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1353
1354                 this_tot->rx_bytes      += bytes;
1355                 this_tot->rx_packets    += packets;
1356         }
1357 }
1358
1359 static void netvsc_get_stats64(struct net_device *net,
1360                                struct rtnl_link_stats64 *t)
1361 {
1362         struct net_device_context *ndev_ctx = netdev_priv(net);
1363         struct netvsc_device *nvdev;
1364         struct netvsc_vf_pcpu_stats vf_tot;
1365         int i;
1366
1367         rcu_read_lock();
1368
1369         nvdev = rcu_dereference(ndev_ctx->nvdev);
1370         if (!nvdev)
1371                 goto out;
1372
1373         netdev_stats_to_stats64(t, &net->stats);
1374
1375         netvsc_get_vf_stats(net, &vf_tot);
1376         t->rx_packets += vf_tot.rx_packets;
1377         t->tx_packets += vf_tot.tx_packets;
1378         t->rx_bytes   += vf_tot.rx_bytes;
1379         t->tx_bytes   += vf_tot.tx_bytes;
1380         t->tx_dropped += vf_tot.tx_dropped;
1381
1382         for (i = 0; i < nvdev->num_chn; i++) {
1383                 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1384                 const struct netvsc_stats *stats;
1385                 u64 packets, bytes, multicast;
1386                 unsigned int start;
1387
1388                 stats = &nvchan->tx_stats;
1389                 do {
1390                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1391                         packets = stats->packets;
1392                         bytes = stats->bytes;
1393                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1394
1395                 t->tx_bytes     += bytes;
1396                 t->tx_packets   += packets;
1397
1398                 stats = &nvchan->rx_stats;
1399                 do {
1400                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1401                         packets = stats->packets;
1402                         bytes = stats->bytes;
1403                         multicast = stats->multicast + stats->broadcast;
1404                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1405
1406                 t->rx_bytes     += bytes;
1407                 t->rx_packets   += packets;
1408                 t->multicast    += multicast;
1409         }
1410 out:
1411         rcu_read_unlock();
1412 }
1413
1414 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
1415 {
1416         struct net_device_context *ndc = netdev_priv(ndev);
1417         struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1418         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1419         struct sockaddr *addr = p;
1420         int err;
1421
1422         err = eth_prepare_mac_addr_change(ndev, p);
1423         if (err)
1424                 return err;
1425
1426         if (!nvdev)
1427                 return -ENODEV;
1428
1429         if (vf_netdev) {
1430                 err = dev_set_mac_address(vf_netdev, addr, NULL);
1431                 if (err)
1432                         return err;
1433         }
1434
1435         err = rndis_filter_set_device_mac(nvdev, addr->sa_data);
1436         if (!err) {
1437                 eth_commit_mac_addr_change(ndev, p);
1438         } else if (vf_netdev) {
1439                 /* rollback change on VF */
1440                 memcpy(addr->sa_data, ndev->dev_addr, ETH_ALEN);
1441                 dev_set_mac_address(vf_netdev, addr, NULL);
1442         }
1443
1444         return err;
1445 }
1446
1447 static const struct {
1448         char name[ETH_GSTRING_LEN];
1449         u16 offset;
1450 } netvsc_stats[] = {
1451         { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
1452         { "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
1453         { "tx_no_space",  offsetof(struct netvsc_ethtool_stats, tx_no_space) },
1454         { "tx_too_big",   offsetof(struct netvsc_ethtool_stats, tx_too_big) },
1455         { "tx_busy",      offsetof(struct netvsc_ethtool_stats, tx_busy) },
1456         { "tx_send_full", offsetof(struct netvsc_ethtool_stats, tx_send_full) },
1457         { "rx_comp_busy", offsetof(struct netvsc_ethtool_stats, rx_comp_busy) },
1458         { "rx_no_memory", offsetof(struct netvsc_ethtool_stats, rx_no_memory) },
1459         { "stop_queue", offsetof(struct netvsc_ethtool_stats, stop_queue) },
1460         { "wake_queue", offsetof(struct netvsc_ethtool_stats, wake_queue) },
1461         { "vlan_error", offsetof(struct netvsc_ethtool_stats, vlan_error) },
1462 }, pcpu_stats[] = {
1463         { "cpu%u_rx_packets",
1464                 offsetof(struct netvsc_ethtool_pcpu_stats, rx_packets) },
1465         { "cpu%u_rx_bytes",
1466                 offsetof(struct netvsc_ethtool_pcpu_stats, rx_bytes) },
1467         { "cpu%u_tx_packets",
1468                 offsetof(struct netvsc_ethtool_pcpu_stats, tx_packets) },
1469         { "cpu%u_tx_bytes",
1470                 offsetof(struct netvsc_ethtool_pcpu_stats, tx_bytes) },
1471         { "cpu%u_vf_rx_packets",
1472                 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_packets) },
1473         { "cpu%u_vf_rx_bytes",
1474                 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_bytes) },
1475         { "cpu%u_vf_tx_packets",
1476                 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_packets) },
1477         { "cpu%u_vf_tx_bytes",
1478                 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_bytes) },
1479 }, vf_stats[] = {
1480         { "vf_rx_packets", offsetof(struct netvsc_vf_pcpu_stats, rx_packets) },
1481         { "vf_rx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, rx_bytes) },
1482         { "vf_tx_packets", offsetof(struct netvsc_vf_pcpu_stats, tx_packets) },
1483         { "vf_tx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, tx_bytes) },
1484         { "vf_tx_dropped", offsetof(struct netvsc_vf_pcpu_stats, tx_dropped) },
1485 };
1486
1487 #define NETVSC_GLOBAL_STATS_LEN ARRAY_SIZE(netvsc_stats)
1488 #define NETVSC_VF_STATS_LEN     ARRAY_SIZE(vf_stats)
1489
1490 /* statistics per queue (rx/tx packets/bytes) */
1491 #define NETVSC_PCPU_STATS_LEN (num_present_cpus() * ARRAY_SIZE(pcpu_stats))
1492
1493 /* 5 statistics per queue (rx/tx packets/bytes, rx xdp_drop) */
1494 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 5)
1495
1496 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
1497 {
1498         struct net_device_context *ndc = netdev_priv(dev);
1499         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1500
1501         if (!nvdev)
1502                 return -ENODEV;
1503
1504         switch (string_set) {
1505         case ETH_SS_STATS:
1506                 return NETVSC_GLOBAL_STATS_LEN
1507                         + NETVSC_VF_STATS_LEN
1508                         + NETVSC_QUEUE_STATS_LEN(nvdev)
1509                         + NETVSC_PCPU_STATS_LEN;
1510         default:
1511                 return -EINVAL;
1512         }
1513 }
1514
1515 static void netvsc_get_ethtool_stats(struct net_device *dev,
1516                                      struct ethtool_stats *stats, u64 *data)
1517 {
1518         struct net_device_context *ndc = netdev_priv(dev);
1519         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1520         const void *nds = &ndc->eth_stats;
1521         const struct netvsc_stats *qstats;
1522         struct netvsc_vf_pcpu_stats sum;
1523         struct netvsc_ethtool_pcpu_stats *pcpu_sum;
1524         unsigned int start;
1525         u64 packets, bytes;
1526         u64 xdp_drop;
1527         int i, j, cpu;
1528
1529         if (!nvdev)
1530                 return;
1531
1532         for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
1533                 data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1534
1535         netvsc_get_vf_stats(dev, &sum);
1536         for (j = 0; j < NETVSC_VF_STATS_LEN; j++)
1537                 data[i++] = *(u64 *)((void *)&sum + vf_stats[j].offset);
1538
1539         for (j = 0; j < nvdev->num_chn; j++) {
1540                 qstats = &nvdev->chan_table[j].tx_stats;
1541
1542                 do {
1543                         start = u64_stats_fetch_begin_irq(&qstats->syncp);
1544                         packets = qstats->packets;
1545                         bytes = qstats->bytes;
1546                 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1547                 data[i++] = packets;
1548                 data[i++] = bytes;
1549
1550                 qstats = &nvdev->chan_table[j].rx_stats;
1551                 do {
1552                         start = u64_stats_fetch_begin_irq(&qstats->syncp);
1553                         packets = qstats->packets;
1554                         bytes = qstats->bytes;
1555                         xdp_drop = qstats->xdp_drop;
1556                 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1557                 data[i++] = packets;
1558                 data[i++] = bytes;
1559                 data[i++] = xdp_drop;
1560         }
1561
1562         pcpu_sum = kvmalloc_array(num_possible_cpus(),
1563                                   sizeof(struct netvsc_ethtool_pcpu_stats),
1564                                   GFP_KERNEL);
1565         netvsc_get_pcpu_stats(dev, pcpu_sum);
1566         for_each_present_cpu(cpu) {
1567                 struct netvsc_ethtool_pcpu_stats *this_sum = &pcpu_sum[cpu];
1568
1569                 for (j = 0; j < ARRAY_SIZE(pcpu_stats); j++)
1570                         data[i++] = *(u64 *)((void *)this_sum
1571                                              + pcpu_stats[j].offset);
1572         }
1573         kvfree(pcpu_sum);
1574 }
1575
1576 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1577 {
1578         struct net_device_context *ndc = netdev_priv(dev);
1579         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1580         u8 *p = data;
1581         int i, cpu;
1582
1583         if (!nvdev)
1584                 return;
1585
1586         switch (stringset) {
1587         case ETH_SS_STATS:
1588                 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++) {
1589                         memcpy(p, netvsc_stats[i].name, ETH_GSTRING_LEN);
1590                         p += ETH_GSTRING_LEN;
1591                 }
1592
1593                 for (i = 0; i < ARRAY_SIZE(vf_stats); i++) {
1594                         memcpy(p, vf_stats[i].name, ETH_GSTRING_LEN);
1595                         p += ETH_GSTRING_LEN;
1596                 }
1597
1598                 for (i = 0; i < nvdev->num_chn; i++) {
1599                         sprintf(p, "tx_queue_%u_packets", i);
1600                         p += ETH_GSTRING_LEN;
1601                         sprintf(p, "tx_queue_%u_bytes", i);
1602                         p += ETH_GSTRING_LEN;
1603                         sprintf(p, "rx_queue_%u_packets", i);
1604                         p += ETH_GSTRING_LEN;
1605                         sprintf(p, "rx_queue_%u_bytes", i);
1606                         p += ETH_GSTRING_LEN;
1607                         sprintf(p, "rx_queue_%u_xdp_drop", i);
1608                         p += ETH_GSTRING_LEN;
1609                 }
1610
1611                 for_each_present_cpu(cpu) {
1612                         for (i = 0; i < ARRAY_SIZE(pcpu_stats); i++) {
1613                                 sprintf(p, pcpu_stats[i].name, cpu);
1614                                 p += ETH_GSTRING_LEN;
1615                         }
1616                 }
1617
1618                 break;
1619         }
1620 }
1621
1622 static int
1623 netvsc_get_rss_hash_opts(struct net_device_context *ndc,
1624                          struct ethtool_rxnfc *info)
1625 {
1626         const u32 l4_flag = RXH_L4_B_0_1 | RXH_L4_B_2_3;
1627
1628         info->data = RXH_IP_SRC | RXH_IP_DST;
1629
1630         switch (info->flow_type) {
1631         case TCP_V4_FLOW:
1632                 if (ndc->l4_hash & HV_TCP4_L4HASH)
1633                         info->data |= l4_flag;
1634
1635                 break;
1636
1637         case TCP_V6_FLOW:
1638                 if (ndc->l4_hash & HV_TCP6_L4HASH)
1639                         info->data |= l4_flag;
1640
1641                 break;
1642
1643         case UDP_V4_FLOW:
1644                 if (ndc->l4_hash & HV_UDP4_L4HASH)
1645                         info->data |= l4_flag;
1646
1647                 break;
1648
1649         case UDP_V6_FLOW:
1650                 if (ndc->l4_hash & HV_UDP6_L4HASH)
1651                         info->data |= l4_flag;
1652
1653                 break;
1654
1655         case IPV4_FLOW:
1656         case IPV6_FLOW:
1657                 break;
1658         default:
1659                 info->data = 0;
1660                 break;
1661         }
1662
1663         return 0;
1664 }
1665
1666 static int
1667 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1668                  u32 *rules)
1669 {
1670         struct net_device_context *ndc = netdev_priv(dev);
1671         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1672
1673         if (!nvdev)
1674                 return -ENODEV;
1675
1676         switch (info->cmd) {
1677         case ETHTOOL_GRXRINGS:
1678                 info->data = nvdev->num_chn;
1679                 return 0;
1680
1681         case ETHTOOL_GRXFH:
1682                 return netvsc_get_rss_hash_opts(ndc, info);
1683         }
1684         return -EOPNOTSUPP;
1685 }
1686
1687 static int netvsc_set_rss_hash_opts(struct net_device_context *ndc,
1688                                     struct ethtool_rxnfc *info)
1689 {
1690         if (info->data == (RXH_IP_SRC | RXH_IP_DST |
1691                            RXH_L4_B_0_1 | RXH_L4_B_2_3)) {
1692                 switch (info->flow_type) {
1693                 case TCP_V4_FLOW:
1694                         ndc->l4_hash |= HV_TCP4_L4HASH;
1695                         break;
1696
1697                 case TCP_V6_FLOW:
1698                         ndc->l4_hash |= HV_TCP6_L4HASH;
1699                         break;
1700
1701                 case UDP_V4_FLOW:
1702                         ndc->l4_hash |= HV_UDP4_L4HASH;
1703                         break;
1704
1705                 case UDP_V6_FLOW:
1706                         ndc->l4_hash |= HV_UDP6_L4HASH;
1707                         break;
1708
1709                 default:
1710                         return -EOPNOTSUPP;
1711                 }
1712
1713                 return 0;
1714         }
1715
1716         if (info->data == (RXH_IP_SRC | RXH_IP_DST)) {
1717                 switch (info->flow_type) {
1718                 case TCP_V4_FLOW:
1719                         ndc->l4_hash &= ~HV_TCP4_L4HASH;
1720                         break;
1721
1722                 case TCP_V6_FLOW:
1723                         ndc->l4_hash &= ~HV_TCP6_L4HASH;
1724                         break;
1725
1726                 case UDP_V4_FLOW:
1727                         ndc->l4_hash &= ~HV_UDP4_L4HASH;
1728                         break;
1729
1730                 case UDP_V6_FLOW:
1731                         ndc->l4_hash &= ~HV_UDP6_L4HASH;
1732                         break;
1733
1734                 default:
1735                         return -EOPNOTSUPP;
1736                 }
1737
1738                 return 0;
1739         }
1740
1741         return -EOPNOTSUPP;
1742 }
1743
1744 static int
1745 netvsc_set_rxnfc(struct net_device *ndev, struct ethtool_rxnfc *info)
1746 {
1747         struct net_device_context *ndc = netdev_priv(ndev);
1748
1749         if (info->cmd == ETHTOOL_SRXFH)
1750                 return netvsc_set_rss_hash_opts(ndc, info);
1751
1752         return -EOPNOTSUPP;
1753 }
1754
1755 static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1756 {
1757         return NETVSC_HASH_KEYLEN;
1758 }
1759
1760 static u32 netvsc_rss_indir_size(struct net_device *dev)
1761 {
1762         return ITAB_NUM;
1763 }
1764
1765 static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key,
1766                            u8 *hfunc)
1767 {
1768         struct net_device_context *ndc = netdev_priv(dev);
1769         struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1770         struct rndis_device *rndis_dev;
1771         int i;
1772
1773         if (!ndev)
1774                 return -ENODEV;
1775
1776         if (hfunc)
1777                 *hfunc = ETH_RSS_HASH_TOP;      /* Toeplitz */
1778
1779         rndis_dev = ndev->extension;
1780         if (indir) {
1781                 for (i = 0; i < ITAB_NUM; i++)
1782                         indir[i] = ndc->rx_table[i];
1783         }
1784
1785         if (key)
1786                 memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1787
1788         return 0;
1789 }
1790
1791 static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
1792                            const u8 *key, const u8 hfunc)
1793 {
1794         struct net_device_context *ndc = netdev_priv(dev);
1795         struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1796         struct rndis_device *rndis_dev;
1797         int i;
1798
1799         if (!ndev)
1800                 return -ENODEV;
1801
1802         if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
1803                 return -EOPNOTSUPP;
1804
1805         rndis_dev = ndev->extension;
1806         if (indir) {
1807                 for (i = 0; i < ITAB_NUM; i++)
1808                         if (indir[i] >= ndev->num_chn)
1809                                 return -EINVAL;
1810
1811                 for (i = 0; i < ITAB_NUM; i++)
1812                         ndc->rx_table[i] = indir[i];
1813         }
1814
1815         if (!key) {
1816                 if (!indir)
1817                         return 0;
1818
1819                 key = rndis_dev->rss_key;
1820         }
1821
1822         return rndis_filter_set_rss_param(rndis_dev, key);
1823 }
1824
1825 /* Hyper-V RNDIS protocol does not have ring in the HW sense.
1826  * It does have pre-allocated receive area which is divided into sections.
1827  */
1828 static void __netvsc_get_ringparam(struct netvsc_device *nvdev,
1829                                    struct ethtool_ringparam *ring)
1830 {
1831         u32 max_buf_size;
1832
1833         ring->rx_pending = nvdev->recv_section_cnt;
1834         ring->tx_pending = nvdev->send_section_cnt;
1835
1836         if (nvdev->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
1837                 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
1838         else
1839                 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
1840
1841         ring->rx_max_pending = max_buf_size / nvdev->recv_section_size;
1842         ring->tx_max_pending = NETVSC_SEND_BUFFER_SIZE
1843                 / nvdev->send_section_size;
1844 }
1845
1846 static void netvsc_get_ringparam(struct net_device *ndev,
1847                                  struct ethtool_ringparam *ring)
1848 {
1849         struct net_device_context *ndevctx = netdev_priv(ndev);
1850         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1851
1852         if (!nvdev)
1853                 return;
1854
1855         __netvsc_get_ringparam(nvdev, ring);
1856 }
1857
1858 static int netvsc_set_ringparam(struct net_device *ndev,
1859                                 struct ethtool_ringparam *ring)
1860 {
1861         struct net_device_context *ndevctx = netdev_priv(ndev);
1862         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1863         struct netvsc_device_info *device_info;
1864         struct ethtool_ringparam orig;
1865         u32 new_tx, new_rx;
1866         int ret = 0;
1867
1868         if (!nvdev || nvdev->destroy)
1869                 return -ENODEV;
1870
1871         memset(&orig, 0, sizeof(orig));
1872         __netvsc_get_ringparam(nvdev, &orig);
1873
1874         new_tx = clamp_t(u32, ring->tx_pending,
1875                          NETVSC_MIN_TX_SECTIONS, orig.tx_max_pending);
1876         new_rx = clamp_t(u32, ring->rx_pending,
1877                          NETVSC_MIN_RX_SECTIONS, orig.rx_max_pending);
1878
1879         if (new_tx == orig.tx_pending &&
1880             new_rx == orig.rx_pending)
1881                 return 0;        /* no change */
1882
1883         device_info = netvsc_devinfo_get(nvdev);
1884
1885         if (!device_info)
1886                 return -ENOMEM;
1887
1888         device_info->send_sections = new_tx;
1889         device_info->recv_sections = new_rx;
1890
1891         ret = netvsc_detach(ndev, nvdev);
1892         if (ret)
1893                 goto out;
1894
1895         ret = netvsc_attach(ndev, device_info);
1896         if (ret) {
1897                 device_info->send_sections = orig.tx_pending;
1898                 device_info->recv_sections = orig.rx_pending;
1899
1900                 if (netvsc_attach(ndev, device_info))
1901                         netdev_err(ndev, "restoring ringparam failed");
1902         }
1903
1904 out:
1905         netvsc_devinfo_put(device_info);
1906         return ret;
1907 }
1908
1909 static netdev_features_t netvsc_fix_features(struct net_device *ndev,
1910                                              netdev_features_t features)
1911 {
1912         struct net_device_context *ndevctx = netdev_priv(ndev);
1913         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1914
1915         if (!nvdev || nvdev->destroy)
1916                 return features;
1917
1918         if ((features & NETIF_F_LRO) && netvsc_xdp_get(nvdev)) {
1919                 features ^= NETIF_F_LRO;
1920                 netdev_info(ndev, "Skip LRO - unsupported with XDP\n");
1921         }
1922
1923         return features;
1924 }
1925
1926 static int netvsc_set_features(struct net_device *ndev,
1927                                netdev_features_t features)
1928 {
1929         netdev_features_t change = features ^ ndev->features;
1930         struct net_device_context *ndevctx = netdev_priv(ndev);
1931         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1932         struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1933         struct ndis_offload_params offloads;
1934         int ret = 0;
1935
1936         if (!nvdev || nvdev->destroy)
1937                 return -ENODEV;
1938
1939         if (!(change & NETIF_F_LRO))
1940                 goto syncvf;
1941
1942         memset(&offloads, 0, sizeof(struct ndis_offload_params));
1943
1944         if (features & NETIF_F_LRO) {
1945                 offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1946                 offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1947         } else {
1948                 offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1949                 offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1950         }
1951
1952         ret = rndis_filter_set_offload_params(ndev, nvdev, &offloads);
1953
1954         if (ret) {
1955                 features ^= NETIF_F_LRO;
1956                 ndev->features = features;
1957         }
1958
1959 syncvf:
1960         if (!vf_netdev)
1961                 return ret;
1962
1963         vf_netdev->wanted_features = features;
1964         netdev_update_features(vf_netdev);
1965
1966         return ret;
1967 }
1968
1969 static int netvsc_get_regs_len(struct net_device *netdev)
1970 {
1971         return VRSS_SEND_TAB_SIZE * sizeof(u32);
1972 }
1973
1974 static void netvsc_get_regs(struct net_device *netdev,
1975                             struct ethtool_regs *regs, void *p)
1976 {
1977         struct net_device_context *ndc = netdev_priv(netdev);
1978         u32 *regs_buff = p;
1979
1980         /* increase the version, if buffer format is changed. */
1981         regs->version = 1;
1982
1983         memcpy(regs_buff, ndc->tx_table, VRSS_SEND_TAB_SIZE * sizeof(u32));
1984 }
1985
1986 static u32 netvsc_get_msglevel(struct net_device *ndev)
1987 {
1988         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1989
1990         return ndev_ctx->msg_enable;
1991 }
1992
1993 static void netvsc_set_msglevel(struct net_device *ndev, u32 val)
1994 {
1995         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1996
1997         ndev_ctx->msg_enable = val;
1998 }
1999
2000 static const struct ethtool_ops ethtool_ops = {
2001         .get_drvinfo    = netvsc_get_drvinfo,
2002         .get_regs_len   = netvsc_get_regs_len,
2003         .get_regs       = netvsc_get_regs,
2004         .get_msglevel   = netvsc_get_msglevel,
2005         .set_msglevel   = netvsc_set_msglevel,
2006         .get_link       = ethtool_op_get_link,
2007         .get_ethtool_stats = netvsc_get_ethtool_stats,
2008         .get_sset_count = netvsc_get_sset_count,
2009         .get_strings    = netvsc_get_strings,
2010         .get_channels   = netvsc_get_channels,
2011         .set_channels   = netvsc_set_channels,
2012         .get_ts_info    = ethtool_op_get_ts_info,
2013         .get_rxnfc      = netvsc_get_rxnfc,
2014         .set_rxnfc      = netvsc_set_rxnfc,
2015         .get_rxfh_key_size = netvsc_get_rxfh_key_size,
2016         .get_rxfh_indir_size = netvsc_rss_indir_size,
2017         .get_rxfh       = netvsc_get_rxfh,
2018         .set_rxfh       = netvsc_set_rxfh,
2019         .get_link_ksettings = netvsc_get_link_ksettings,
2020         .set_link_ksettings = netvsc_set_link_ksettings,
2021         .get_ringparam  = netvsc_get_ringparam,
2022         .set_ringparam  = netvsc_set_ringparam,
2023 };
2024
2025 static const struct net_device_ops device_ops = {
2026         .ndo_open =                     netvsc_open,
2027         .ndo_stop =                     netvsc_close,
2028         .ndo_start_xmit =               netvsc_start_xmit,
2029         .ndo_change_rx_flags =          netvsc_change_rx_flags,
2030         .ndo_set_rx_mode =              netvsc_set_rx_mode,
2031         .ndo_fix_features =             netvsc_fix_features,
2032         .ndo_set_features =             netvsc_set_features,
2033         .ndo_change_mtu =               netvsc_change_mtu,
2034         .ndo_validate_addr =            eth_validate_addr,
2035         .ndo_set_mac_address =          netvsc_set_mac_addr,
2036         .ndo_select_queue =             netvsc_select_queue,
2037         .ndo_get_stats64 =              netvsc_get_stats64,
2038         .ndo_bpf =                      netvsc_bpf,
2039 };
2040
2041 /*
2042  * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
2043  * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
2044  * present send GARP packet to network peers with netif_notify_peers().
2045  */
2046 static void netvsc_link_change(struct work_struct *w)
2047 {
2048         struct net_device_context *ndev_ctx =
2049                 container_of(w, struct net_device_context, dwork.work);
2050         struct hv_device *device_obj = ndev_ctx->device_ctx;
2051         struct net_device *net = hv_get_drvdata(device_obj);
2052         struct netvsc_device *net_device;
2053         struct rndis_device *rdev;
2054         struct netvsc_reconfig *event = NULL;
2055         bool notify = false, reschedule = false;
2056         unsigned long flags, next_reconfig, delay;
2057
2058         /* if changes are happening, comeback later */
2059         if (!rtnl_trylock()) {
2060                 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
2061                 return;
2062         }
2063
2064         net_device = rtnl_dereference(ndev_ctx->nvdev);
2065         if (!net_device)
2066                 goto out_unlock;
2067
2068         rdev = net_device->extension;
2069
2070         next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
2071         if (time_is_after_jiffies(next_reconfig)) {
2072                 /* link_watch only sends one notification with current state
2073                  * per second, avoid doing reconfig more frequently. Handle
2074                  * wrap around.
2075                  */
2076                 delay = next_reconfig - jiffies;
2077                 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
2078                 schedule_delayed_work(&ndev_ctx->dwork, delay);
2079                 goto out_unlock;
2080         }
2081         ndev_ctx->last_reconfig = jiffies;
2082
2083         spin_lock_irqsave(&ndev_ctx->lock, flags);
2084         if (!list_empty(&ndev_ctx->reconfig_events)) {
2085                 event = list_first_entry(&ndev_ctx->reconfig_events,
2086                                          struct netvsc_reconfig, list);
2087                 list_del(&event->list);
2088                 reschedule = !list_empty(&ndev_ctx->reconfig_events);
2089         }
2090         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
2091
2092         if (!event)
2093                 goto out_unlock;
2094
2095         switch (event->event) {
2096                 /* Only the following events are possible due to the check in
2097                  * netvsc_linkstatus_callback()
2098                  */
2099         case RNDIS_STATUS_MEDIA_CONNECT:
2100                 if (rdev->link_state) {
2101                         rdev->link_state = false;
2102                         netif_carrier_on(net);
2103                         netvsc_tx_enable(net_device, net);
2104                 } else {
2105                         notify = true;
2106                 }
2107                 kfree(event);
2108                 break;
2109         case RNDIS_STATUS_MEDIA_DISCONNECT:
2110                 if (!rdev->link_state) {
2111                         rdev->link_state = true;
2112                         netif_carrier_off(net);
2113                         netvsc_tx_disable(net_device, net);
2114                 }
2115                 kfree(event);
2116                 break;
2117         case RNDIS_STATUS_NETWORK_CHANGE:
2118                 /* Only makes sense if carrier is present */
2119                 if (!rdev->link_state) {
2120                         rdev->link_state = true;
2121                         netif_carrier_off(net);
2122                         netvsc_tx_disable(net_device, net);
2123                         event->event = RNDIS_STATUS_MEDIA_CONNECT;
2124                         spin_lock_irqsave(&ndev_ctx->lock, flags);
2125                         list_add(&event->list, &ndev_ctx->reconfig_events);
2126                         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
2127                         reschedule = true;
2128                 }
2129                 break;
2130         }
2131
2132         rtnl_unlock();
2133
2134         if (notify)
2135                 netdev_notify_peers(net);
2136
2137         /* link_watch only sends one notification with current state per
2138          * second, handle next reconfig event in 2 seconds.
2139          */
2140         if (reschedule)
2141                 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
2142
2143         return;
2144
2145 out_unlock:
2146         rtnl_unlock();
2147 }
2148
2149 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
2150 {
2151         struct net_device_context *net_device_ctx;
2152         struct net_device *dev;
2153
2154         dev = netdev_master_upper_dev_get(vf_netdev);
2155         if (!dev || dev->netdev_ops != &device_ops)
2156                 return NULL;    /* not a netvsc device */
2157
2158         net_device_ctx = netdev_priv(dev);
2159         if (!rtnl_dereference(net_device_ctx->nvdev))
2160                 return NULL;    /* device is removed */
2161
2162         return dev;
2163 }
2164
2165 /* Called when VF is injecting data into network stack.
2166  * Change the associated network device from VF to netvsc.
2167  * note: already called with rcu_read_lock
2168  */
2169 static rx_handler_result_t netvsc_vf_handle_frame(struct sk_buff **pskb)
2170 {
2171         struct sk_buff *skb = *pskb;
2172         struct net_device *ndev = rcu_dereference(skb->dev->rx_handler_data);
2173         struct net_device_context *ndev_ctx = netdev_priv(ndev);
2174         struct netvsc_vf_pcpu_stats *pcpu_stats
2175                  = this_cpu_ptr(ndev_ctx->vf_stats);
2176
2177         skb = skb_share_check(skb, GFP_ATOMIC);
2178         if (unlikely(!skb))
2179                 return RX_HANDLER_CONSUMED;
2180
2181         *pskb = skb;
2182
2183         skb->dev = ndev;
2184
2185         u64_stats_update_begin(&pcpu_stats->syncp);
2186         pcpu_stats->rx_packets++;
2187         pcpu_stats->rx_bytes += skb->len;
2188         u64_stats_update_end(&pcpu_stats->syncp);
2189
2190         return RX_HANDLER_ANOTHER;
2191 }
2192
2193 static int netvsc_vf_join(struct net_device *vf_netdev,
2194                           struct net_device *ndev)
2195 {
2196         struct net_device_context *ndev_ctx = netdev_priv(ndev);
2197         int ret;
2198
2199         ret = netdev_rx_handler_register(vf_netdev,
2200                                          netvsc_vf_handle_frame, ndev);
2201         if (ret != 0) {
2202                 netdev_err(vf_netdev,
2203                            "can not register netvsc VF receive handler (err = %d)\n",
2204                            ret);
2205                 goto rx_handler_failed;
2206         }
2207
2208         ret = netdev_master_upper_dev_link(vf_netdev, ndev,
2209                                            NULL, NULL, NULL);
2210         if (ret != 0) {
2211                 netdev_err(vf_netdev,
2212                            "can not set master device %s (err = %d)\n",
2213                            ndev->name, ret);
2214                 goto upper_link_failed;
2215         }
2216
2217         /* set slave flag before open to prevent IPv6 addrconf */
2218         vf_netdev->flags |= IFF_SLAVE;
2219
2220         schedule_delayed_work(&ndev_ctx->vf_takeover, VF_TAKEOVER_INT);
2221
2222         call_netdevice_notifiers(NETDEV_JOIN, vf_netdev);
2223
2224         netdev_info(vf_netdev, "joined to %s\n", ndev->name);
2225         return 0;
2226
2227 upper_link_failed:
2228         netdev_rx_handler_unregister(vf_netdev);
2229 rx_handler_failed:
2230         return ret;
2231 }
2232
2233 static void __netvsc_vf_setup(struct net_device *ndev,
2234                               struct net_device *vf_netdev)
2235 {
2236         int ret;
2237
2238         /* Align MTU of VF with master */
2239         ret = dev_set_mtu(vf_netdev, ndev->mtu);
2240         if (ret)
2241                 netdev_warn(vf_netdev,
2242                             "unable to change mtu to %u\n", ndev->mtu);
2243
2244         /* set multicast etc flags on VF */
2245         dev_change_flags(vf_netdev, ndev->flags | IFF_SLAVE, NULL);
2246
2247         /* sync address list from ndev to VF */
2248         netif_addr_lock_bh(ndev);
2249         dev_uc_sync(vf_netdev, ndev);
2250         dev_mc_sync(vf_netdev, ndev);
2251         netif_addr_unlock_bh(ndev);
2252
2253         if (netif_running(ndev)) {
2254                 ret = dev_open(vf_netdev, NULL);
2255                 if (ret)
2256                         netdev_warn(vf_netdev,
2257                                     "unable to open: %d\n", ret);
2258         }
2259 }
2260
2261 /* Setup VF as slave of the synthetic device.
2262  * Runs in workqueue to avoid recursion in netlink callbacks.
2263  */
2264 static void netvsc_vf_setup(struct work_struct *w)
2265 {
2266         struct net_device_context *ndev_ctx
2267                 = container_of(w, struct net_device_context, vf_takeover.work);
2268         struct net_device *ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2269         struct net_device *vf_netdev;
2270
2271         if (!rtnl_trylock()) {
2272                 schedule_delayed_work(&ndev_ctx->vf_takeover, 0);
2273                 return;
2274         }
2275
2276         vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2277         if (vf_netdev)
2278                 __netvsc_vf_setup(ndev, vf_netdev);
2279
2280         rtnl_unlock();
2281 }
2282
2283 /* Find netvsc by VF serial number.
2284  * The PCI hyperv controller records the serial number as the slot kobj name.
2285  */
2286 static struct net_device *get_netvsc_byslot(const struct net_device *vf_netdev)
2287 {
2288         struct device *parent = vf_netdev->dev.parent;
2289         struct net_device_context *ndev_ctx;
2290         struct pci_dev *pdev;
2291         u32 serial;
2292
2293         if (!parent || !dev_is_pci(parent))
2294                 return NULL; /* not a PCI device */
2295
2296         pdev = to_pci_dev(parent);
2297         if (!pdev->slot) {
2298                 netdev_notice(vf_netdev, "no PCI slot information\n");
2299                 return NULL;
2300         }
2301
2302         if (kstrtou32(pci_slot_name(pdev->slot), 10, &serial)) {
2303                 netdev_notice(vf_netdev, "Invalid vf serial:%s\n",
2304                               pci_slot_name(pdev->slot));
2305                 return NULL;
2306         }
2307
2308         list_for_each_entry(ndev_ctx, &netvsc_dev_list, list) {
2309                 if (!ndev_ctx->vf_alloc)
2310                         continue;
2311
2312                 if (ndev_ctx->vf_serial == serial)
2313                         return hv_get_drvdata(ndev_ctx->device_ctx);
2314         }
2315
2316         netdev_notice(vf_netdev,
2317                       "no netdev found for vf serial:%u\n", serial);
2318         return NULL;
2319 }
2320
2321 static int netvsc_register_vf(struct net_device *vf_netdev)
2322 {
2323         struct net_device_context *net_device_ctx;
2324         struct netvsc_device *netvsc_dev;
2325         struct bpf_prog *prog;
2326         struct net_device *ndev;
2327         int ret;
2328
2329         if (vf_netdev->addr_len != ETH_ALEN)
2330                 return NOTIFY_DONE;
2331
2332         ndev = get_netvsc_byslot(vf_netdev);
2333         if (!ndev)
2334                 return NOTIFY_DONE;
2335
2336         net_device_ctx = netdev_priv(ndev);
2337         netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2338         if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
2339                 return NOTIFY_DONE;
2340
2341         /* if synthetic interface is a different namespace,
2342          * then move the VF to that namespace; join will be
2343          * done again in that context.
2344          */
2345         if (!net_eq(dev_net(ndev), dev_net(vf_netdev))) {
2346                 ret = dev_change_net_namespace(vf_netdev,
2347                                                dev_net(ndev), "eth%d");
2348                 if (ret)
2349                         netdev_err(vf_netdev,
2350                                    "could not move to same namespace as %s: %d\n",
2351                                    ndev->name, ret);
2352                 else
2353                         netdev_info(vf_netdev,
2354                                     "VF moved to namespace with: %s\n",
2355                                     ndev->name);
2356                 return NOTIFY_DONE;
2357         }
2358
2359         netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
2360
2361         if (netvsc_vf_join(vf_netdev, ndev) != 0)
2362                 return NOTIFY_DONE;
2363
2364         dev_hold(vf_netdev);
2365         rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
2366
2367         vf_netdev->wanted_features = ndev->features;
2368         netdev_update_features(vf_netdev);
2369
2370         prog = netvsc_xdp_get(netvsc_dev);
2371         netvsc_vf_setxdp(vf_netdev, prog);
2372
2373         return NOTIFY_OK;
2374 }
2375
2376 /* Change the data path when VF UP/DOWN/CHANGE are detected.
2377  *
2378  * Typically a UP or DOWN event is followed by a CHANGE event, so
2379  * net_device_ctx->data_path_is_vf is used to cache the current data path
2380  * to avoid the duplicate call of netvsc_switch_datapath() and the duplicate
2381  * message.
2382  *
2383  * During hibernation, if a VF NIC driver (e.g. mlx5) preserves the network
2384  * interface, there is only the CHANGE event and no UP or DOWN event.
2385  */
2386 static int netvsc_vf_changed(struct net_device *vf_netdev)
2387 {
2388         struct net_device_context *net_device_ctx;
2389         struct netvsc_device *netvsc_dev;
2390         struct net_device *ndev;
2391         bool vf_is_up = netif_running(vf_netdev);
2392
2393         ndev = get_netvsc_byref(vf_netdev);
2394         if (!ndev)
2395                 return NOTIFY_DONE;
2396
2397         net_device_ctx = netdev_priv(ndev);
2398         netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2399         if (!netvsc_dev)
2400                 return NOTIFY_DONE;
2401
2402         if (net_device_ctx->data_path_is_vf == vf_is_up)
2403                 return NOTIFY_OK;
2404         net_device_ctx->data_path_is_vf = vf_is_up;
2405
2406         netvsc_switch_datapath(ndev, vf_is_up);
2407         netdev_info(ndev, "Data path switched %s VF: %s\n",
2408                     vf_is_up ? "to" : "from", vf_netdev->name);
2409
2410         return NOTIFY_OK;
2411 }
2412
2413 static int netvsc_unregister_vf(struct net_device *vf_netdev)
2414 {
2415         struct net_device *ndev;
2416         struct net_device_context *net_device_ctx;
2417
2418         ndev = get_netvsc_byref(vf_netdev);
2419         if (!ndev)
2420                 return NOTIFY_DONE;
2421
2422         net_device_ctx = netdev_priv(ndev);
2423         cancel_delayed_work_sync(&net_device_ctx->vf_takeover);
2424
2425         netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
2426
2427         netvsc_vf_setxdp(vf_netdev, NULL);
2428
2429         netdev_rx_handler_unregister(vf_netdev);
2430         netdev_upper_dev_unlink(vf_netdev, ndev);
2431         RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
2432         dev_put(vf_netdev);
2433
2434         return NOTIFY_OK;
2435 }
2436
2437 static int netvsc_probe(struct hv_device *dev,
2438                         const struct hv_vmbus_device_id *dev_id)
2439 {
2440         struct net_device *net = NULL;
2441         struct net_device_context *net_device_ctx;
2442         struct netvsc_device_info *device_info = NULL;
2443         struct netvsc_device *nvdev;
2444         int ret = -ENOMEM;
2445
2446         net = alloc_etherdev_mq(sizeof(struct net_device_context),
2447                                 VRSS_CHANNEL_MAX);
2448         if (!net)
2449                 goto no_net;
2450
2451         netif_carrier_off(net);
2452
2453         netvsc_init_settings(net);
2454
2455         net_device_ctx = netdev_priv(net);
2456         net_device_ctx->device_ctx = dev;
2457         net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
2458         if (netif_msg_probe(net_device_ctx))
2459                 netdev_dbg(net, "netvsc msg_enable: %d\n",
2460                            net_device_ctx->msg_enable);
2461
2462         hv_set_drvdata(dev, net);
2463
2464         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
2465
2466         spin_lock_init(&net_device_ctx->lock);
2467         INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
2468         INIT_DELAYED_WORK(&net_device_ctx->vf_takeover, netvsc_vf_setup);
2469
2470         net_device_ctx->vf_stats
2471                 = netdev_alloc_pcpu_stats(struct netvsc_vf_pcpu_stats);
2472         if (!net_device_ctx->vf_stats)
2473                 goto no_stats;
2474
2475         net->netdev_ops = &device_ops;
2476         net->ethtool_ops = &ethtool_ops;
2477         SET_NETDEV_DEV(net, &dev->device);
2478
2479         /* We always need headroom for rndis header */
2480         net->needed_headroom = RNDIS_AND_PPI_SIZE;
2481
2482         /* Initialize the number of queues to be 1, we may change it if more
2483          * channels are offered later.
2484          */
2485         netif_set_real_num_tx_queues(net, 1);
2486         netif_set_real_num_rx_queues(net, 1);
2487
2488         /* Notify the netvsc driver of the new device */
2489         device_info = netvsc_devinfo_get(NULL);
2490
2491         if (!device_info) {
2492                 ret = -ENOMEM;
2493                 goto devinfo_failed;
2494         }
2495
2496         nvdev = rndis_filter_device_add(dev, device_info);
2497         if (IS_ERR(nvdev)) {
2498                 ret = PTR_ERR(nvdev);
2499                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
2500                 goto rndis_failed;
2501         }
2502
2503         memcpy(net->dev_addr, device_info->mac_adr, ETH_ALEN);
2504
2505         /* We must get rtnl lock before scheduling nvdev->subchan_work,
2506          * otherwise netvsc_subchan_work() can get rtnl lock first and wait
2507          * all subchannels to show up, but that may not happen because
2508          * netvsc_probe() can't get rtnl lock and as a result vmbus_onoffer()
2509          * -> ... -> device_add() -> ... -> __device_attach() can't get
2510          * the device lock, so all the subchannels can't be processed --
2511          * finally netvsc_subchan_work() hangs forever.
2512          */
2513         rtnl_lock();
2514
2515         if (nvdev->num_chn > 1)
2516                 schedule_work(&nvdev->subchan_work);
2517
2518         /* hw_features computed in rndis_netdev_set_hwcaps() */
2519         net->features = net->hw_features |
2520                 NETIF_F_HIGHDMA | NETIF_F_HW_VLAN_CTAG_TX |
2521                 NETIF_F_HW_VLAN_CTAG_RX;
2522         net->vlan_features = net->features;
2523
2524         netdev_lockdep_set_classes(net);
2525
2526         /* MTU range: 68 - 1500 or 65521 */
2527         net->min_mtu = NETVSC_MTU_MIN;
2528         if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
2529                 net->max_mtu = NETVSC_MTU - ETH_HLEN;
2530         else
2531                 net->max_mtu = ETH_DATA_LEN;
2532
2533         nvdev->tx_disable = false;
2534
2535         ret = register_netdevice(net);
2536         if (ret != 0) {
2537                 pr_err("Unable to register netdev.\n");
2538                 goto register_failed;
2539         }
2540
2541         list_add(&net_device_ctx->list, &netvsc_dev_list);
2542         rtnl_unlock();
2543
2544         netvsc_devinfo_put(device_info);
2545         return 0;
2546
2547 register_failed:
2548         rtnl_unlock();
2549         rndis_filter_device_remove(dev, nvdev);
2550 rndis_failed:
2551         netvsc_devinfo_put(device_info);
2552 devinfo_failed:
2553         free_percpu(net_device_ctx->vf_stats);
2554 no_stats:
2555         hv_set_drvdata(dev, NULL);
2556         free_netdev(net);
2557 no_net:
2558         return ret;
2559 }
2560
2561 static int netvsc_remove(struct hv_device *dev)
2562 {
2563         struct net_device_context *ndev_ctx;
2564         struct net_device *vf_netdev, *net;
2565         struct netvsc_device *nvdev;
2566
2567         net = hv_get_drvdata(dev);
2568         if (net == NULL) {
2569                 dev_err(&dev->device, "No net device to remove\n");
2570                 return 0;
2571         }
2572
2573         ndev_ctx = netdev_priv(net);
2574
2575         cancel_delayed_work_sync(&ndev_ctx->dwork);
2576
2577         rtnl_lock();
2578         nvdev = rtnl_dereference(ndev_ctx->nvdev);
2579         if (nvdev) {
2580                 cancel_work_sync(&nvdev->subchan_work);
2581                 netvsc_xdp_set(net, NULL, NULL, nvdev);
2582         }
2583
2584         /*
2585          * Call to the vsc driver to let it know that the device is being
2586          * removed. Also blocks mtu and channel changes.
2587          */
2588         vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2589         if (vf_netdev)
2590                 netvsc_unregister_vf(vf_netdev);
2591
2592         if (nvdev)
2593                 rndis_filter_device_remove(dev, nvdev);
2594
2595         unregister_netdevice(net);
2596         list_del(&ndev_ctx->list);
2597
2598         rtnl_unlock();
2599
2600         hv_set_drvdata(dev, NULL);
2601
2602         free_percpu(ndev_ctx->vf_stats);
2603         free_netdev(net);
2604         return 0;
2605 }
2606
2607 static int netvsc_suspend(struct hv_device *dev)
2608 {
2609         struct net_device_context *ndev_ctx;
2610         struct netvsc_device *nvdev;
2611         struct net_device *net;
2612         int ret;
2613
2614         net = hv_get_drvdata(dev);
2615
2616         ndev_ctx = netdev_priv(net);
2617         cancel_delayed_work_sync(&ndev_ctx->dwork);
2618
2619         rtnl_lock();
2620
2621         nvdev = rtnl_dereference(ndev_ctx->nvdev);
2622         if (nvdev == NULL) {
2623                 ret = -ENODEV;
2624                 goto out;
2625         }
2626
2627         /* Save the current config info */
2628         ndev_ctx->saved_netvsc_dev_info = netvsc_devinfo_get(nvdev);
2629
2630         ret = netvsc_detach(net, nvdev);
2631 out:
2632         rtnl_unlock();
2633
2634         return ret;
2635 }
2636
2637 static int netvsc_resume(struct hv_device *dev)
2638 {
2639         struct net_device *net = hv_get_drvdata(dev);
2640         struct net_device_context *net_device_ctx;
2641         struct netvsc_device_info *device_info;
2642         int ret;
2643
2644         rtnl_lock();
2645
2646         net_device_ctx = netdev_priv(net);
2647
2648         /* Reset the data path to the netvsc NIC before re-opening the vmbus
2649          * channel. Later netvsc_netdev_event() will switch the data path to
2650          * the VF upon the UP or CHANGE event.
2651          */
2652         net_device_ctx->data_path_is_vf = false;
2653         device_info = net_device_ctx->saved_netvsc_dev_info;
2654
2655         ret = netvsc_attach(net, device_info);
2656
2657         netvsc_devinfo_put(device_info);
2658         net_device_ctx->saved_netvsc_dev_info = NULL;
2659
2660         rtnl_unlock();
2661
2662         return ret;
2663 }
2664 static const struct hv_vmbus_device_id id_table[] = {
2665         /* Network guid */
2666         { HV_NIC_GUID, },
2667         { },
2668 };
2669
2670 MODULE_DEVICE_TABLE(vmbus, id_table);
2671
2672 /* The one and only one */
2673 static struct  hv_driver netvsc_drv = {
2674         .name = KBUILD_MODNAME,
2675         .id_table = id_table,
2676         .probe = netvsc_probe,
2677         .remove = netvsc_remove,
2678         .suspend = netvsc_suspend,
2679         .resume = netvsc_resume,
2680         .driver = {
2681                 .probe_type = PROBE_FORCE_SYNCHRONOUS,
2682         },
2683 };
2684
2685 /*
2686  * On Hyper-V, every VF interface is matched with a corresponding
2687  * synthetic interface. The synthetic interface is presented first
2688  * to the guest. When the corresponding VF instance is registered,
2689  * we will take care of switching the data path.
2690  */
2691 static int netvsc_netdev_event(struct notifier_block *this,
2692                                unsigned long event, void *ptr)
2693 {
2694         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2695
2696         /* Skip our own events */
2697         if (event_dev->netdev_ops == &device_ops)
2698                 return NOTIFY_DONE;
2699
2700         /* Avoid non-Ethernet type devices */
2701         if (event_dev->type != ARPHRD_ETHER)
2702                 return NOTIFY_DONE;
2703
2704         /* Avoid Vlan dev with same MAC registering as VF */
2705         if (is_vlan_dev(event_dev))
2706                 return NOTIFY_DONE;
2707
2708         /* Avoid Bonding master dev with same MAC registering as VF */
2709         if ((event_dev->priv_flags & IFF_BONDING) &&
2710             (event_dev->flags & IFF_MASTER))
2711                 return NOTIFY_DONE;
2712
2713         switch (event) {
2714         case NETDEV_REGISTER:
2715                 return netvsc_register_vf(event_dev);
2716         case NETDEV_UNREGISTER:
2717                 return netvsc_unregister_vf(event_dev);
2718         case NETDEV_UP:
2719         case NETDEV_DOWN:
2720         case NETDEV_CHANGE:
2721                 return netvsc_vf_changed(event_dev);
2722         default:
2723                 return NOTIFY_DONE;
2724         }
2725 }
2726
2727 static struct notifier_block netvsc_netdev_notifier = {
2728         .notifier_call = netvsc_netdev_event,
2729 };
2730
2731 static void __exit netvsc_drv_exit(void)
2732 {
2733         unregister_netdevice_notifier(&netvsc_netdev_notifier);
2734         vmbus_driver_unregister(&netvsc_drv);
2735 }
2736
2737 static int __init netvsc_drv_init(void)
2738 {
2739         int ret;
2740
2741         if (ring_size < RING_SIZE_MIN) {
2742                 ring_size = RING_SIZE_MIN;
2743                 pr_info("Increased ring_size to %u (min allowed)\n",
2744                         ring_size);
2745         }
2746         netvsc_ring_bytes = ring_size * PAGE_SIZE;
2747
2748         ret = vmbus_driver_register(&netvsc_drv);
2749         if (ret)
2750                 return ret;
2751
2752         register_netdevice_notifier(&netvsc_netdev_notifier);
2753         return 0;
2754 }
2755
2756 MODULE_LICENSE("GPL");
2757 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
2758
2759 module_init(netvsc_drv_init);
2760 module_exit(netvsc_drv_exit);