drivers/net/hyperv/netvsc_drv.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (c) 2009, Microsoft Corporation.
   4  *
   5  * Authors:
   6  *   Haiyang Zhang <haiyangz@microsoft.com>
   7  *   Hank Janssen  <hjanssen@microsoft.com>
   8  */
   9 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  10
  11 #include <linux/init.h>
  12 #include <linux/atomic.h>
  13 #include <linux/ethtool.h>
  14 #include <linux/module.h>
  15 #include <linux/highmem.h>
  16 #include <linux/device.h>
  17 #include <linux/io.h>
  18 #include <linux/delay.h>
  19 #include <linux/netdevice.h>
  20 #include <linux/inetdevice.h>
  21 #include <linux/etherdevice.h>
  22 #include <linux/pci.h>
  23 #include <linux/skbuff.h>
  24 #include <linux/if_vlan.h>
  25 #include <linux/in.h>
  26 #include <linux/slab.h>
  27 #include <linux/rtnetlink.h>
  28 #include <linux/netpoll.h>
  29 #include <linux/bpf.h>
  30
  31 #include <net/arp.h>
  32 #include <net/route.h>
  33 #include <net/sock.h>
  34 #include <net/pkt_sched.h>
  35 #include <net/checksum.h>
  36 #include <net/ip6_checksum.h>
  37
  38 #include "hyperv_net.h"
  39
  40 #define RING_SIZE_MIN   64
  41 #define RETRY_US_LO     5000
  42 #define RETRY_US_HI     10000
  43 #define RETRY_MAX       2000    /* >10 sec */
  44
  45 #define LINKCHANGE_INT (2 * HZ)
  46 #define VF_TAKEOVER_INT (HZ / 10)
  47
  48 static unsigned int ring_size __ro_after_init = 128;
  49 module_param(ring_size, uint, 0444);
  50 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
  51 unsigned int netvsc_ring_bytes __ro_after_init;
  52
  53 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
  54                                 NETIF_MSG_LINK | NETIF_MSG_IFUP |
  55                                 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
  56                                 NETIF_MSG_TX_ERR;
  57
  58 static int debug = -1;
  59 module_param(debug, int, 0444);
  60 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  61
  62 static LIST_HEAD(netvsc_dev_list);
  63
  64 static void netvsc_change_rx_flags(struct net_device *net, int change)
  65 {
  66         struct net_device_context *ndev_ctx = netdev_priv(net);
  67         struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
  68         int inc;
  69
  70         if (!vf_netdev)
  71                 return;
  72
  73         if (change & IFF_PROMISC) {
  74                 inc = (net->flags & IFF_PROMISC) ? 1 : -1;
  75                 dev_set_promiscuity(vf_netdev, inc);
  76         }
  77
  78         if (change & IFF_ALLMULTI) {
  79                 inc = (net->flags & IFF_ALLMULTI) ? 1 : -1;
  80                 dev_set_allmulti(vf_netdev, inc);
  81         }
  82 }
  83
  84 static void netvsc_set_rx_mode(struct net_device *net)
  85 {
  86         struct net_device_context *ndev_ctx = netdev_priv(net);
  87         struct net_device *vf_netdev;
  88         struct netvsc_device *nvdev;
  89
  90         rcu_read_lock();
  91         vf_netdev = rcu_dereference(ndev_ctx->vf_netdev);
  92         if (vf_netdev) {
  93                 dev_uc_sync(vf_netdev, net);
  94                 dev_mc_sync(vf_netdev, net);
  95         }
  96
  97         nvdev = rcu_dereference(ndev_ctx->nvdev);
  98         if (nvdev)
  99                 rndis_filter_update(nvdev);
 100         rcu_read_unlock();
 101 }
 102
 103 static void netvsc_tx_enable(struct netvsc_device *nvscdev,
 104                              struct net_device *ndev)
 105 {
 106         nvscdev->tx_disable = false;
 107         virt_wmb(); /* ensure queue wake up mechanism is on */
 108
 109         netif_tx_wake_all_queues(ndev);
 110 }
 111
 112 static int netvsc_open(struct net_device *net)
 113 {
 114         struct net_device_context *ndev_ctx = netdev_priv(net);
 115         struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
 116         struct netvsc_device *nvdev = rtnl_dereference(ndev_ctx->nvdev);
 117         struct rndis_device *rdev;
 118         int ret = 0;
 119
 120         netif_carrier_off(net);
 121
 122         /* Open up the device */
 123         ret = rndis_filter_open(nvdev);
 124         if (ret != 0) {
 125                 netdev_err(net, "unable to open device (ret %d).\n", ret);
 126                 return ret;
 127         }
 128
 129         rdev = nvdev->extension;
 130         if (!rdev->link_state) {
 131                 netif_carrier_on(net);
 132                 netvsc_tx_enable(nvdev, net);
 133         }
 134
 135         if (vf_netdev) {
 136                 /* Setting synthetic device up transparently sets
 137                  * slave as up. If open fails, then slave will be
 138                  * still be offline (and not used).
 139                  */
 140                 ret = dev_open(vf_netdev, NULL);
 141                 if (ret)
 142                         netdev_warn(net,
 143                                     "unable to open slave: %s: %d\n",
 144                                     vf_netdev->name, ret);
 145         }
 146         return 0;
 147 }
 148
 149 static int netvsc_wait_until_empty(struct netvsc_device *nvdev)
 150 {
 151         unsigned int retry = 0;
 152         int i;
 153
 154         /* Ensure pending bytes in ring are read */
 155         for (;;) {
 156                 u32 aread = 0;
 157
 158                 for (i = 0; i < nvdev->num_chn; i++) {
 159                         struct vmbus_channel *chn
 160                                 = nvdev->chan_table[i].channel;
 161
 162                         if (!chn)
 163                                 continue;
 164
 165                         /* make sure receive not running now */
 166                         napi_synchronize(&nvdev->chan_table[i].napi);
 167
 168                         aread = hv_get_bytes_to_read(&chn->inbound);
 169                         if (aread)
 170                                 break;
 171
 172                         aread = hv_get_bytes_to_read(&chn->outbound);
 173                         if (aread)
 174                                 break;
 175                 }
 176
 177                 if (aread == 0)
 178                         return 0;
 179
 180                 if (++retry > RETRY_MAX)
 181                         return -ETIMEDOUT;
 182
 183                 usleep_range(RETRY_US_LO, RETRY_US_HI);
 184         }
 185 }
 186
 187 static void netvsc_tx_disable(struct netvsc_device *nvscdev,
 188                               struct net_device *ndev)
 189 {
 190         if (nvscdev) {
 191                 nvscdev->tx_disable = true;
 192                 virt_wmb(); /* ensure txq will not wake up after stop */
 193         }
 194
 195         netif_tx_disable(ndev);
 196 }
 197
 198 static int netvsc_close(struct net_device *net)
 199 {
 200         struct net_device_context *net_device_ctx = netdev_priv(net);
 201         struct net_device *vf_netdev
 202                 = rtnl_dereference(net_device_ctx->vf_netdev);
 203         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
 204         int ret;
 205
 206         netvsc_tx_disable(nvdev, net);
 207
 208         /* No need to close rndis filter if it is removed already */
 209         if (!nvdev)
 210                 return 0;
 211
 212         ret = rndis_filter_close(nvdev);
 213         if (ret != 0) {
 214                 netdev_err(net, "unable to close device (ret %d).\n", ret);
 215                 return ret;
 216         }
 217
 218         ret = netvsc_wait_until_empty(nvdev);
 219         if (ret)
 220                 netdev_err(net, "Ring buffer not empty after closing rndis\n");
 221
 222         if (vf_netdev)
 223                 dev_close(vf_netdev);
 224
 225         return ret;
 226 }
 227
 228 static inline void *init_ppi_data(struct rndis_message *msg,
 229                                   u32 ppi_size, u32 pkt_type)
 230 {
 231         struct rndis_packet *rndis_pkt = &msg->msg.pkt;
 232         struct rndis_per_packet_info *ppi;
 233
 234         rndis_pkt->data_offset += ppi_size;
 235         ppi = (void *)rndis_pkt + rndis_pkt->per_pkt_info_offset
 236                 + rndis_pkt->per_pkt_info_len;
 237
 238         ppi->size = ppi_size;
 239         ppi->type = pkt_type;
 240         ppi->internal = 0;
 241         ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
 242
 243         rndis_pkt->per_pkt_info_len += ppi_size;
 244
 245         return ppi + 1;
 246 }
 247
 248 /* Azure hosts don't support non-TCP port numbers in hashing for fragmented
 249  * packets. We can use ethtool to change UDP hash level when necessary.
 250  */
 251 static inline u32 netvsc_get_hash(
 252         struct sk_buff *skb,
 253         const struct net_device_context *ndc)
 254 {
 255         struct flow_keys flow;
 256         u32 hash, pkt_proto = 0;
 257         static u32 hashrnd __read_mostly;
 258
 259         net_get_random_once(&hashrnd, sizeof(hashrnd));
 260
 261         if (!skb_flow_dissect_flow_keys(skb, &flow, 0))
 262                 return 0;
 263
 264         switch (flow.basic.ip_proto) {
 265         case IPPROTO_TCP:
 266                 if (flow.basic.n_proto == htons(ETH_P_IP))
 267                         pkt_proto = HV_TCP4_L4HASH;
 268                 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
 269                         pkt_proto = HV_TCP6_L4HASH;
 270
 271                 break;
 272
 273         case IPPROTO_UDP:
 274                 if (flow.basic.n_proto == htons(ETH_P_IP))
 275                         pkt_proto = HV_UDP4_L4HASH;
 276                 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
 277                         pkt_proto = HV_UDP6_L4HASH;
 278
 279                 break;
 280         }
 281
 282         if (pkt_proto & ndc->l4_hash) {
 283                 return skb_get_hash(skb);
 284         } else {
 285                 if (flow.basic.n_proto == htons(ETH_P_IP))
 286                         hash = jhash2((u32 *)&flow.addrs.v4addrs, 2, hashrnd);
 287                 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
 288                         hash = jhash2((u32 *)&flow.addrs.v6addrs, 8, hashrnd);
 289                 else
 290                         return 0;
 291
 292                 __skb_set_sw_hash(skb, hash, false);
 293         }
 294
 295         return hash;
 296 }
 297
 298 static inline int netvsc_get_tx_queue(struct net_device *ndev,
 299                                       struct sk_buff *skb, int old_idx)
 300 {
 301         const struct net_device_context *ndc = netdev_priv(ndev);
 302         struct sock *sk = skb->sk;
 303         int q_idx;
 304
 305         q_idx = ndc->tx_table[netvsc_get_hash(skb, ndc) &
 306                               (VRSS_SEND_TAB_SIZE - 1)];
 307
 308         /* If queue index changed record the new value */
 309         if (q_idx != old_idx &&
 310             sk && sk_fullsock(sk) && rcu_access_pointer(sk->sk_dst_cache))
 311                 sk_tx_queue_set(sk, q_idx);
 312
 313         return q_idx;
 314 }
 315
 316 /*
 317  * Select queue for transmit.
 318  *
 319  * If a valid queue has already been assigned, then use that.
 320  * Otherwise compute tx queue based on hash and the send table.
 321  *
 322  * This is basically similar to default (netdev_pick_tx) with the added step
 323  * of using the host send_table when no other queue has been assigned.
 324  *
 325  * TODO support XPS - but get_xps_queue not exported
 326  */
 327 static u16 netvsc_pick_tx(struct net_device *ndev, struct sk_buff *skb)
 328 {
 329         int q_idx = sk_tx_queue_get(skb->sk);
 330
 331         if (q_idx < 0 || skb->ooo_okay || q_idx >= ndev->real_num_tx_queues) {
 332                 /* If forwarding a packet, we use the recorded queue when
 333                  * available for better cache locality.
 334                  */
 335                 if (skb_rx_queue_recorded(skb))
 336                         q_idx = skb_get_rx_queue(skb);
 337                 else
 338                         q_idx = netvsc_get_tx_queue(ndev, skb, q_idx);
 339         }
 340
 341         return q_idx;
 342 }
 343
 344 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
 345                                struct net_device *sb_dev)
 346 {
 347         struct net_device_context *ndc = netdev_priv(ndev);
 348         struct net_device *vf_netdev;
 349         u16 txq;
 350
 351         rcu_read_lock();
 352         vf_netdev = rcu_dereference(ndc->vf_netdev);
 353         if (vf_netdev) {
 354                 const struct net_device_ops *vf_ops = vf_netdev->netdev_ops;
 355
 356                 if (vf_ops->ndo_select_queue)
 357                         txq = vf_ops->ndo_select_queue(vf_netdev, skb, sb_dev);
 358                 else
 359                         txq = netdev_pick_tx(vf_netdev, skb, NULL);
 360
 361                 /* Record the queue selected by VF so that it can be
 362                  * used for common case where VF has more queues than
 363                  * the synthetic device.
 364                  */
 365                 qdisc_skb_cb(skb)->slave_dev_queue_mapping = txq;
 366         } else {
 367                 txq = netvsc_pick_tx(ndev, skb);
 368         }
 369         rcu_read_unlock();
 370
 371         while (txq >= ndev->real_num_tx_queues)
 372                 txq -= ndev->real_num_tx_queues;
 373
 374         return txq;
 375 }
 376
 377 static u32 fill_pg_buf(unsigned long hvpfn, u32 offset, u32 len,
 378                        struct hv_page_buffer *pb)
 379 {
 380         int j = 0;
 381
 382         hvpfn += offset >> HV_HYP_PAGE_SHIFT;
 383         offset = offset & ~HV_HYP_PAGE_MASK;
 384
 385         while (len > 0) {
 386                 unsigned long bytes;
 387
 388                 bytes = HV_HYP_PAGE_SIZE - offset;
 389                 if (bytes > len)
 390                         bytes = len;
 391                 pb[j].pfn = hvpfn;
 392                 pb[j].offset = offset;
 393                 pb[j].len = bytes;
 394
 395                 offset += bytes;
 396                 len -= bytes;
 397
 398                 if (offset == HV_HYP_PAGE_SIZE && len) {
 399                         hvpfn++;
 400                         offset = 0;
 401                         j++;
 402                 }
 403         }
 404
 405         return j + 1;
 406 }
 407
 408 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
 409                            struct hv_netvsc_packet *packet,
 410                            struct hv_page_buffer *pb)
 411 {
 412         u32 slots_used = 0;
 413         char *data = skb->data;
 414         int frags = skb_shinfo(skb)->nr_frags;
 415         int i;
 416
 417         /* The packet is laid out thus:
 418          * 1. hdr: RNDIS header and PPI
 419          * 2. skb linear data
 420          * 3. skb fragment data
 421          */
 422         slots_used += fill_pg_buf(virt_to_hvpfn(hdr),
 423                                   offset_in_hvpage(hdr),
 424                                   len,
 425                                   &pb[slots_used]);
 426
 427         packet->rmsg_size = len;
 428         packet->rmsg_pgcnt = slots_used;
 429
 430         slots_used += fill_pg_buf(virt_to_hvpfn(data),
 431                                   offset_in_hvpage(data),
 432                                   skb_headlen(skb),
 433                                   &pb[slots_used]);
 434
 435         for (i = 0; i < frags; i++) {
 436                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 437
 438                 slots_used += fill_pg_buf(page_to_hvpfn(skb_frag_page(frag)),
 439                                           skb_frag_off(frag),
 440                                           skb_frag_size(frag),
 441                                           &pb[slots_used]);
 442         }
 443         return slots_used;
 444 }
 445
 446 static int count_skb_frag_slots(struct sk_buff *skb)
 447 {
 448         int i, frags = skb_shinfo(skb)->nr_frags;
 449         int pages = 0;
 450
 451         for (i = 0; i < frags; i++) {
 452                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 453                 unsigned long size = skb_frag_size(frag);
 454                 unsigned long offset = skb_frag_off(frag);
 455
 456                 /* Skip unused frames from start of page */
 457                 offset &= ~HV_HYP_PAGE_MASK;
 458                 pages += HVPFN_UP(offset + size);
 459         }
 460         return pages;
 461 }
 462
 463 static int netvsc_get_slots(struct sk_buff *skb)
 464 {
 465         char *data = skb->data;
 466         unsigned int offset = offset_in_hvpage(data);
 467         unsigned int len = skb_headlen(skb);
 468         int slots;
 469         int frag_slots;
 470
 471         slots = DIV_ROUND_UP(offset + len, HV_HYP_PAGE_SIZE);
 472         frag_slots = count_skb_frag_slots(skb);
 473         return slots + frag_slots;
 474 }
 475
 476 static u32 net_checksum_info(struct sk_buff *skb)
 477 {
 478         if (skb->protocol == htons(ETH_P_IP)) {
 479                 struct iphdr *ip = ip_hdr(skb);
 480
 481                 if (ip->protocol == IPPROTO_TCP)
 482                         return TRANSPORT_INFO_IPV4_TCP;
 483                 else if (ip->protocol == IPPROTO_UDP)
 484                         return TRANSPORT_INFO_IPV4_UDP;
 485         } else {
 486                 struct ipv6hdr *ip6 = ipv6_hdr(skb);
 487
 488                 if (ip6->nexthdr == IPPROTO_TCP)
 489                         return TRANSPORT_INFO_IPV6_TCP;
 490                 else if (ip6->nexthdr == IPPROTO_UDP)
 491                         return TRANSPORT_INFO_IPV6_UDP;
 492         }
 493
 494         return TRANSPORT_INFO_NOT_IP;
 495 }
 496
 497 /* Send skb on the slave VF device. */
 498 static int netvsc_vf_xmit(struct net_device *net, struct net_device *vf_netdev,
 499                           struct sk_buff *skb)
 500 {
 501         struct net_device_context *ndev_ctx = netdev_priv(net);
 502         unsigned int len = skb->len;
 503         int rc;
 504
 505         skb->dev = vf_netdev;
 506         skb_record_rx_queue(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
 507
 508         rc = dev_queue_xmit(skb);
 509         if (likely(rc == NET_XMIT_SUCCESS || rc == NET_XMIT_CN)) {
 510                 struct netvsc_vf_pcpu_stats *pcpu_stats
 511                         = this_cpu_ptr(ndev_ctx->vf_stats);
 512
 513                 u64_stats_update_begin(&pcpu_stats->syncp);
 514                 pcpu_stats->tx_packets++;
 515                 pcpu_stats->tx_bytes += len;
 516                 u64_stats_update_end(&pcpu_stats->syncp);
 517         } else {
 518                 this_cpu_inc(ndev_ctx->vf_stats->tx_dropped);
 519         }
 520
 521         return rc;
 522 }
 523
 524 static int netvsc_xmit(struct sk_buff *skb, struct net_device *net, bool xdp_tx)
 525 {
 526         struct net_device_context *net_device_ctx = netdev_priv(net);
 527         struct hv_netvsc_packet *packet = NULL;
 528         int ret;
 529         unsigned int num_data_pgs;
 530         struct rndis_message *rndis_msg;
 531         struct net_device *vf_netdev;
 532         u32 rndis_msg_size;
 533         u32 hash;
 534         struct hv_page_buffer pb[MAX_PAGE_BUFFER_COUNT];
 535
 536         /* If VF is present and up then redirect packets to it.
 537          * Skip the VF if it is marked down or has no carrier.
 538          * If netpoll is in uses, then VF can not be used either.
 539          */
 540         vf_netdev = rcu_dereference_bh(net_device_ctx->vf_netdev);
 541         if (vf_netdev && netif_running(vf_netdev) &&
 542             netif_carrier_ok(vf_netdev) && !netpoll_tx_running(net) &&
 543             net_device_ctx->data_path_is_vf)
 544                 return netvsc_vf_xmit(net, vf_netdev, skb);
 545
 546         /* We will atmost need two pages to describe the rndis
 547          * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
 548          * of pages in a single packet. If skb is scattered around
 549          * more pages we try linearizing it.
 550          */
 551
 552         num_data_pgs = netvsc_get_slots(skb) + 2;
 553
 554         if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
 555                 ++net_device_ctx->eth_stats.tx_scattered;
 556
 557                 if (skb_linearize(skb))
 558                         goto no_memory;
 559
 560                 num_data_pgs = netvsc_get_slots(skb) + 2;
 561                 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
 562                         ++net_device_ctx->eth_stats.tx_too_big;
 563                         goto drop;
 564                 }
 565         }
 566
 567         /*
 568          * Place the rndis header in the skb head room and
 569          * the skb->cb will be used for hv_netvsc_packet
 570          * structure.
 571          */
 572         ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
 573         if (ret)
 574                 goto no_memory;
 575
 576         /* Use the skb control buffer for building up the packet */
 577         BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
 578                         sizeof_field(struct sk_buff, cb));
 579         packet = (struct hv_netvsc_packet *)skb->cb;
 580
 581         packet->q_idx = skb_get_queue_mapping(skb);
 582
 583         packet->total_data_buflen = skb->len;
 584         packet->total_bytes = skb->len;
 585         packet->total_packets = 1;
 586
 587         rndis_msg = (struct rndis_message *)skb->head;
 588
 589         /* Add the rndis header */
 590         rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
 591         rndis_msg->msg_len = packet->total_data_buflen;
 592
 593         rndis_msg->msg.pkt = (struct rndis_packet) {
 594                 .data_offset = sizeof(struct rndis_packet),
 595                 .data_len = packet->total_data_buflen,
 596                 .per_pkt_info_offset = sizeof(struct rndis_packet),
 597         };
 598
 599         rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
 600
 601         hash = skb_get_hash_raw(skb);
 602         if (hash != 0 && net->real_num_tx_queues > 1) {
 603                 u32 *hash_info;
 604
 605                 rndis_msg_size += NDIS_HASH_PPI_SIZE;
 606                 hash_info = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
 607                                           NBL_HASH_VALUE);
 608                 *hash_info = hash;
 609         }
 610
 611         /* When using AF_PACKET we need to drop VLAN header from
 612          * the frame and update the SKB to allow the HOST OS
 613          * to transmit the 802.1Q packet
 614          */
 615         if (skb->protocol == htons(ETH_P_8021Q)) {
 616                 u16 vlan_tci;
 617
 618                 skb_reset_mac_header(skb);
 619                 if (eth_type_vlan(eth_hdr(skb)->h_proto)) {
 620                         if (unlikely(__skb_vlan_pop(skb, &vlan_tci) != 0)) {
 621                                 ++net_device_ctx->eth_stats.vlan_error;
 622                                 goto drop;
 623                         }
 624
 625                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci);
 626                         /* Update the NDIS header pkt lengths */
 627                         packet->total_data_buflen -= VLAN_HLEN;
 628                         packet->total_bytes -= VLAN_HLEN;
 629                         rndis_msg->msg_len = packet->total_data_buflen;
 630                         rndis_msg->msg.pkt.data_len = packet->total_data_buflen;
 631                 }
 632         }
 633
 634         if (skb_vlan_tag_present(skb)) {
 635                 struct ndis_pkt_8021q_info *vlan;
 636
 637                 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
 638                 vlan = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
 639                                      IEEE_8021Q_INFO);
 640
 641                 vlan->value = 0;
 642                 vlan->vlanid = skb_vlan_tag_get_id(skb);
 643                 vlan->cfi = skb_vlan_tag_get_cfi(skb);
 644                 vlan->pri = skb_vlan_tag_get_prio(skb);
 645         }
 646
 647         if (skb_is_gso(skb)) {
 648                 struct ndis_tcp_lso_info *lso_info;
 649
 650                 rndis_msg_size += NDIS_LSO_PPI_SIZE;
 651                 lso_info = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
 652                                          TCP_LARGESEND_PKTINFO);
 653
 654                 lso_info->value = 0;
 655                 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
 656                 if (skb->protocol == htons(ETH_P_IP)) {
 657                         lso_info->lso_v2_transmit.ip_version =
 658                                 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
 659                         ip_hdr(skb)->tot_len = 0;
 660                         ip_hdr(skb)->check = 0;
 661                         tcp_hdr(skb)->check =
 662                                 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
 663                                                    ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 664                 } else {
 665                         lso_info->lso_v2_transmit.ip_version =
 666                                 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
 667                         tcp_v6_gso_csum_prep(skb);
 668                 }
 669                 lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb);
 670                 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
 671         } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
 672                 if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) {
 673                         struct ndis_tcp_ip_checksum_info *csum_info;
 674
 675                         rndis_msg_size += NDIS_CSUM_PPI_SIZE;
 676                         csum_info = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
 677                                                   TCPIP_CHKSUM_PKTINFO);
 678
 679                         csum_info->value = 0;
 680                         csum_info->transmit.tcp_header_offset = skb_transport_offset(skb);
 681
 682                         if (skb->protocol == htons(ETH_P_IP)) {
 683                                 csum_info->transmit.is_ipv4 = 1;
 684
 685                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
 686                                         csum_info->transmit.tcp_checksum = 1;
 687                                 else
 688                                         csum_info->transmit.udp_checksum = 1;
 689                         } else {
 690                                 csum_info->transmit.is_ipv6 = 1;
 691
 692                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
 693                                         csum_info->transmit.tcp_checksum = 1;
 694                                 else
 695                                         csum_info->transmit.udp_checksum = 1;
 696                         }
 697                 } else {
 698                         /* Can't do offload of this type of checksum */
 699                         if (skb_checksum_help(skb))
 700                                 goto drop;
 701                 }
 702         }
 703
 704         /* Start filling in the page buffers with the rndis hdr */
 705         rndis_msg->msg_len += rndis_msg_size;
 706         packet->total_data_buflen = rndis_msg->msg_len;
 707         packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
 708                                                skb, packet, pb);
 709
 710         /* timestamp packet in software */
 711         skb_tx_timestamp(skb);
 712
 713         ret = netvsc_send(net, packet, rndis_msg, pb, skb, xdp_tx);
 714         if (likely(ret == 0))
 715                 return NETDEV_TX_OK;
 716
 717         if (ret == -EAGAIN) {
 718                 ++net_device_ctx->eth_stats.tx_busy;
 719                 return NETDEV_TX_BUSY;
 720         }
 721
 722         if (ret == -ENOSPC)
 723                 ++net_device_ctx->eth_stats.tx_no_space;
 724
 725 drop:
 726         dev_kfree_skb_any(skb);
 727         net->stats.tx_dropped++;
 728
 729         return NETDEV_TX_OK;
 730
 731 no_memory:
 732         ++net_device_ctx->eth_stats.tx_no_memory;
 733         goto drop;
 734 }
 735
 736 static netdev_tx_t netvsc_start_xmit(struct sk_buff *skb,
 737                                      struct net_device *ndev)
 738 {
 739         return netvsc_xmit(skb, ndev, false);
 740 }
 741
 742 /*
 743  * netvsc_linkstatus_callback - Link up/down notification
 744  */
 745 void netvsc_linkstatus_callback(struct net_device *net,
 746                                 struct rndis_message *resp,
 747                                 void *data)
 748 {
 749         struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
 750         struct net_device_context *ndev_ctx = netdev_priv(net);
 751         struct netvsc_reconfig *event;
 752         unsigned long flags;
 753
 754         /* Ensure the packet is big enough to access its fields */
 755         if (resp->msg_len - RNDIS_HEADER_SIZE < sizeof(struct rndis_indicate_status)) {
 756                 netdev_err(net, "invalid rndis_indicate_status packet, len: %u\n",
 757                            resp->msg_len);
 758                 return;
 759         }
 760
 761         /* Copy the RNDIS indicate status into nvchan->recv_buf */
 762         memcpy(indicate, data + RNDIS_HEADER_SIZE, sizeof(*indicate));
 763
 764         /* Update the physical link speed when changing to another vSwitch */
 765         if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
 766                 u32 speed;
 767
 768                 /* Validate status_buf_offset */
 769                 if (indicate->status_buflen < sizeof(speed) ||
 770                     indicate->status_buf_offset < sizeof(*indicate) ||
 771                     resp->msg_len - RNDIS_HEADER_SIZE < indicate->status_buf_offset ||
 772                     resp->msg_len - RNDIS_HEADER_SIZE - indicate->status_buf_offset
 773                                 < indicate->status_buflen) {
 774                         netdev_err(net, "invalid rndis_indicate_status packet\n");
 775                         return;
 776                 }
 777
 778                 speed = *(u32 *)(data + RNDIS_HEADER_SIZE + indicate->status_buf_offset) / 10000;
 779                 ndev_ctx->speed = speed;
 780                 return;
 781         }
 782
 783         /* Handle these link change statuses below */
 784         if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
 785             indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
 786             indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
 787                 return;
 788
 789         if (net->reg_state != NETREG_REGISTERED)
 790                 return;
 791
 792         event = kzalloc(sizeof(*event), GFP_ATOMIC);
 793         if (!event)
 794                 return;
 795         event->event = indicate->status;
 796
 797         spin_lock_irqsave(&ndev_ctx->lock, flags);
 798         list_add_tail(&event->list, &ndev_ctx->reconfig_events);
 799         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
 800
 801         schedule_delayed_work(&ndev_ctx->dwork, 0);
 802 }
 803
 804 static void netvsc_xdp_xmit(struct sk_buff *skb, struct net_device *ndev)
 805 {
 806         int rc;
 807
 808         skb->queue_mapping = skb_get_rx_queue(skb);
 809         __skb_push(skb, ETH_HLEN);
 810
 811         rc = netvsc_xmit(skb, ndev, true);
 812
 813         if (dev_xmit_complete(rc))
 814                 return;
 815
 816         dev_kfree_skb_any(skb);
 817         ndev->stats.tx_dropped++;
 818 }
 819
 820 static void netvsc_comp_ipcsum(struct sk_buff *skb)
 821 {
 822         struct iphdr *iph = (struct iphdr *)skb->data;
 823
 824         iph->check = 0;
 825         iph->check = ip_fast_csum(iph, iph->ihl);
 826 }
 827
 828 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
 829                                              struct netvsc_channel *nvchan,
 830                                              struct xdp_buff *xdp)
 831 {
 832         struct napi_struct *napi = &nvchan->napi;
 833         const struct ndis_pkt_8021q_info *vlan = &nvchan->rsc.vlan;
 834         const struct ndis_tcp_ip_checksum_info *csum_info =
 835                                                 &nvchan->rsc.csum_info;
 836         const u32 *hash_info = &nvchan->rsc.hash_info;
 837         u8 ppi_flags = nvchan->rsc.ppi_flags;
 838         struct sk_buff *skb;
 839         void *xbuf = xdp->data_hard_start;
 840         int i;
 841
 842         if (xbuf) {
 843                 unsigned int hdroom = xdp->data - xdp->data_hard_start;
 844                 unsigned int xlen = xdp->data_end - xdp->data;
 845                 unsigned int frag_size = xdp->frame_sz;
 846
 847                 skb = build_skb(xbuf, frag_size);
 848
 849                 if (!skb) {
 850                         __free_page(virt_to_page(xbuf));
 851                         return NULL;
 852                 }
 853
 854                 skb_reserve(skb, hdroom);
 855                 skb_put(skb, xlen);
 856                 skb->dev = napi->dev;
 857         } else {
 858                 skb = napi_alloc_skb(napi, nvchan->rsc.pktlen);
 859
 860                 if (!skb)
 861                         return NULL;
 862
 863                 /* Copy to skb. This copy is needed here since the memory
 864                  * pointed by hv_netvsc_packet cannot be deallocated.
 865                  */
 866                 for (i = 0; i < nvchan->rsc.cnt; i++)
 867                         skb_put_data(skb, nvchan->rsc.data[i],
 868                                      nvchan->rsc.len[i]);
 869         }
 870
 871         skb->protocol = eth_type_trans(skb, net);
 872
 873         /* skb is already created with CHECKSUM_NONE */
 874         skb_checksum_none_assert(skb);
 875
 876         /* Incoming packets may have IP header checksum verified by the host.
 877          * They may not have IP header checksum computed after coalescing.
 878          * We compute it here if the flags are set, because on Linux, the IP
 879          * checksum is always checked.
 880          */
 881         if ((ppi_flags & NVSC_RSC_CSUM_INFO) && csum_info->receive.ip_checksum_value_invalid &&
 882             csum_info->receive.ip_checksum_succeeded &&
 883             skb->protocol == htons(ETH_P_IP)) {
 884                 /* Check that there is enough space to hold the IP header. */
 885                 if (skb_headlen(skb) < sizeof(struct iphdr)) {
 886                         kfree_skb(skb);
 887                         return NULL;
 888                 }
 889                 netvsc_comp_ipcsum(skb);
 890         }
 891
 892         /* Do L4 checksum offload if enabled and present. */
 893         if ((ppi_flags & NVSC_RSC_CSUM_INFO) && (net->features & NETIF_F_RXCSUM)) {
 894                 if (csum_info->receive.tcp_checksum_succeeded ||
 895                     csum_info->receive.udp_checksum_succeeded)
 896                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 897         }
 898
 899         if ((ppi_flags & NVSC_RSC_HASH_INFO) && (net->features & NETIF_F_RXHASH))
 900                 skb_set_hash(skb, *hash_info, PKT_HASH_TYPE_L4);
 901
 902         if (ppi_flags & NVSC_RSC_VLAN) {
 903                 u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT) |
 904                         (vlan->cfi ? VLAN_CFI_MASK : 0);
 905
 906                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 907                                        vlan_tci);
 908         }
 909
 910         return skb;
 911 }
 912
 913 /*
 914  * netvsc_recv_callback -  Callback when we receive a packet from the
 915  * "wire" on the specified device.
 916  */
 917 int netvsc_recv_callback(struct net_device *net,
 918                          struct netvsc_device *net_device,
 919                          struct netvsc_channel *nvchan)
 920 {
 921         struct net_device_context *net_device_ctx = netdev_priv(net);
 922         struct vmbus_channel *channel = nvchan->channel;
 923         u16 q_idx = channel->offermsg.offer.sub_channel_index;
 924         struct sk_buff *skb;
 925         struct netvsc_stats *rx_stats = &nvchan->rx_stats;
 926         struct xdp_buff xdp;
 927         u32 act;
 928
 929         if (net->reg_state != NETREG_REGISTERED)
 930                 return NVSP_STAT_FAIL;
 931
 932         act = netvsc_run_xdp(net, nvchan, &xdp);
 933
 934         if (act != XDP_PASS && act != XDP_TX) {
 935                 u64_stats_update_begin(&rx_stats->syncp);
 936                 rx_stats->xdp_drop++;
 937                 u64_stats_update_end(&rx_stats->syncp);
 938
 939                 return NVSP_STAT_SUCCESS; /* consumed by XDP */
 940         }
 941
 942         /* Allocate a skb - TODO direct I/O to pages? */
 943         skb = netvsc_alloc_recv_skb(net, nvchan, &xdp);
 944
 945         if (unlikely(!skb)) {
 946                 ++net_device_ctx->eth_stats.rx_no_memory;
 947                 return NVSP_STAT_FAIL;
 948         }
 949
 950         skb_record_rx_queue(skb, q_idx);
 951
 952         /*
 953          * Even if injecting the packet, record the statistics
 954          * on the synthetic device because modifying the VF device
 955          * statistics will not work correctly.
 956          */
 957         u64_stats_update_begin(&rx_stats->syncp);
 958         rx_stats->packets++;
 959         rx_stats->bytes += nvchan->rsc.pktlen;
 960
 961         if (skb->pkt_type == PACKET_BROADCAST)
 962                 ++rx_stats->broadcast;
 963         else if (skb->pkt_type == PACKET_MULTICAST)
 964                 ++rx_stats->multicast;
 965         u64_stats_update_end(&rx_stats->syncp);
 966
 967         if (act == XDP_TX) {
 968                 netvsc_xdp_xmit(skb, net);
 969                 return NVSP_STAT_SUCCESS;
 970         }
 971
 972         napi_gro_receive(&nvchan->napi, skb);
 973         return NVSP_STAT_SUCCESS;
 974 }
 975
 976 static void netvsc_get_drvinfo(struct net_device *net,
 977                                struct ethtool_drvinfo *info)
 978 {
 979         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
 980         strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
 981 }
 982
 983 static void netvsc_get_channels(struct net_device *net,
 984                                 struct ethtool_channels *channel)
 985 {
 986         struct net_device_context *net_device_ctx = netdev_priv(net);
 987         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
 988
 989         if (nvdev) {
 990                 channel->max_combined   = nvdev->max_chn;
 991                 channel->combined_count = nvdev->num_chn;
 992         }
 993 }
 994
 995 /* Alloc struct netvsc_device_info, and initialize it from either existing
 996  * struct netvsc_device, or from default values.
 997  */
 998 static
 999 struct netvsc_device_info *netvsc_devinfo_get(struct netvsc_device *nvdev)
1000 {
1001         struct netvsc_device_info *dev_info;
1002         struct bpf_prog *prog;
1003
1004         dev_info = kzalloc(sizeof(*dev_info), GFP_ATOMIC);
1005
1006         if (!dev_info)
1007                 return NULL;
1008
1009         if (nvdev) {
1010                 ASSERT_RTNL();
1011
1012                 dev_info->num_chn = nvdev->num_chn;
1013                 dev_info->send_sections = nvdev->send_section_cnt;
1014                 dev_info->send_section_size = nvdev->send_section_size;
1015                 dev_info->recv_sections = nvdev->recv_section_cnt;
1016                 dev_info->recv_section_size = nvdev->recv_section_size;
1017
1018                 memcpy(dev_info->rss_key, nvdev->extension->rss_key,
1019                        NETVSC_HASH_KEYLEN);
1020
1021                 prog = netvsc_xdp_get(nvdev);
1022                 if (prog) {
1023                         bpf_prog_inc(prog);
1024                         dev_info->bprog = prog;
1025                 }
1026         } else {
1027                 dev_info->num_chn = VRSS_CHANNEL_DEFAULT;
1028                 dev_info->send_sections = NETVSC_DEFAULT_TX;
1029                 dev_info->send_section_size = NETVSC_SEND_SECTION_SIZE;
1030                 dev_info->recv_sections = NETVSC_DEFAULT_RX;
1031                 dev_info->recv_section_size = NETVSC_RECV_SECTION_SIZE;
1032         }
1033
1034         return dev_info;
1035 }
1036
1037 /* Free struct netvsc_device_info */
1038 static void netvsc_devinfo_put(struct netvsc_device_info *dev_info)
1039 {
1040         if (dev_info->bprog) {
1041                 ASSERT_RTNL();
1042                 bpf_prog_put(dev_info->bprog);
1043         }
1044
1045         kfree(dev_info);
1046 }
1047
1048 static int netvsc_detach(struct net_device *ndev,
1049                          struct netvsc_device *nvdev)
1050 {
1051         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1052         struct hv_device *hdev = ndev_ctx->device_ctx;
1053         int ret;
1054
1055         /* Don't try continuing to try and setup sub channels */
1056         if (cancel_work_sync(&nvdev->subchan_work))
1057                 nvdev->num_chn = 1;
1058
1059         netvsc_xdp_set(ndev, NULL, NULL, nvdev);
1060
1061         /* If device was up (receiving) then shutdown */
1062         if (netif_running(ndev)) {
1063                 netvsc_tx_disable(nvdev, ndev);
1064
1065                 ret = rndis_filter_close(nvdev);
1066                 if (ret) {
1067                         netdev_err(ndev,
1068                                    "unable to close device (ret %d).\n", ret);
1069                         return ret;
1070                 }
1071
1072                 ret = netvsc_wait_until_empty(nvdev);
1073                 if (ret) {
1074                         netdev_err(ndev,
1075                                    "Ring buffer not empty after closing rndis\n");
1076                         return ret;
1077                 }
1078         }
1079
1080         netif_device_detach(ndev);
1081
1082         rndis_filter_device_remove(hdev, nvdev);
1083
1084         return 0;
1085 }
1086
1087 static int netvsc_attach(struct net_device *ndev,
1088                          struct netvsc_device_info *dev_info)
1089 {
1090         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1091         struct hv_device *hdev = ndev_ctx->device_ctx;
1092         struct netvsc_device *nvdev;
1093         struct rndis_device *rdev;
1094         struct bpf_prog *prog;
1095         int ret = 0;
1096
1097         nvdev = rndis_filter_device_add(hdev, dev_info);
1098         if (IS_ERR(nvdev))
1099                 return PTR_ERR(nvdev);
1100
1101         if (nvdev->num_chn > 1) {
1102                 ret = rndis_set_subchannel(ndev, nvdev, dev_info);
1103
1104                 /* if unavailable, just proceed with one queue */
1105                 if (ret) {
1106                         nvdev->max_chn = 1;
1107                         nvdev->num_chn = 1;
1108                 }
1109         }
1110
1111         prog = dev_info->bprog;
1112         if (prog) {
1113                 bpf_prog_inc(prog);
1114                 ret = netvsc_xdp_set(ndev, prog, NULL, nvdev);
1115                 if (ret) {
1116                         bpf_prog_put(prog);
1117                         goto err1;
1118                 }
1119         }
1120
1121         /* In any case device is now ready */
1122         nvdev->tx_disable = false;
1123         netif_device_attach(ndev);
1124
1125         /* Note: enable and attach happen when sub-channels setup */
1126         netif_carrier_off(ndev);
1127
1128         if (netif_running(ndev)) {
1129                 ret = rndis_filter_open(nvdev);
1130                 if (ret)
1131                         goto err2;
1132
1133                 rdev = nvdev->extension;
1134                 if (!rdev->link_state)
1135                         netif_carrier_on(ndev);
1136         }
1137
1138         return 0;
1139
1140 err2:
1141         netif_device_detach(ndev);
1142
1143 err1:
1144         rndis_filter_device_remove(hdev, nvdev);
1145
1146         return ret;
1147 }
1148
1149 static int netvsc_set_channels(struct net_device *net,
1150                                struct ethtool_channels *channels)
1151 {
1152         struct net_device_context *net_device_ctx = netdev_priv(net);
1153         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
1154         unsigned int orig, count = channels->combined_count;
1155         struct netvsc_device_info *device_info;
1156         int ret;
1157
1158         /* We do not support separate count for rx, tx, or other */
1159         if (count == 0 ||
1160             channels->rx_count || channels->tx_count || channels->other_count)
1161                 return -EINVAL;
1162
1163         if (!nvdev || nvdev->destroy)
1164                 return -ENODEV;
1165
1166         if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
1167                 return -EINVAL;
1168
1169         if (count > nvdev->max_chn)
1170                 return -EINVAL;
1171
1172         orig = nvdev->num_chn;
1173
1174         device_info = netvsc_devinfo_get(nvdev);
1175
1176         if (!device_info)
1177                 return -ENOMEM;
1178
1179         device_info->num_chn = count;
1180
1181         ret = netvsc_detach(net, nvdev);
1182         if (ret)
1183                 goto out;
1184
1185         ret = netvsc_attach(net, device_info);
1186         if (ret) {
1187                 device_info->num_chn = orig;
1188                 if (netvsc_attach(net, device_info))
1189                         netdev_err(net, "restoring channel setting failed\n");
1190         }
1191
1192 out:
1193         netvsc_devinfo_put(device_info);
1194         return ret;
1195 }
1196
1197 static void netvsc_init_settings(struct net_device *dev)
1198 {
1199         struct net_device_context *ndc = netdev_priv(dev);
1200
1201         ndc->l4_hash = HV_DEFAULT_L4HASH;
1202
1203         ndc->speed = SPEED_UNKNOWN;
1204         ndc->duplex = DUPLEX_FULL;
1205
1206         dev->features = NETIF_F_LRO;
1207 }
1208
1209 static int netvsc_get_link_ksettings(struct net_device *dev,
1210                                      struct ethtool_link_ksettings *cmd)
1211 {
1212         struct net_device_context *ndc = netdev_priv(dev);
1213         struct net_device *vf_netdev;
1214
1215         vf_netdev = rtnl_dereference(ndc->vf_netdev);
1216
1217         if (vf_netdev)
1218                 return __ethtool_get_link_ksettings(vf_netdev, cmd);
1219
1220         cmd->base.speed = ndc->speed;
1221         cmd->base.duplex = ndc->duplex;
1222         cmd->base.port = PORT_OTHER;
1223
1224         return 0;
1225 }
1226
1227 static int netvsc_set_link_ksettings(struct net_device *dev,
1228                                      const struct ethtool_link_ksettings *cmd)
1229 {
1230         struct net_device_context *ndc = netdev_priv(dev);
1231         struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1232
1233         if (vf_netdev) {
1234                 if (!vf_netdev->ethtool_ops->set_link_ksettings)
1235                         return -EOPNOTSUPP;
1236
1237                 return vf_netdev->ethtool_ops->set_link_ksettings(vf_netdev,
1238                                                                   cmd);
1239         }
1240
1241         return ethtool_virtdev_set_link_ksettings(dev, cmd,
1242                                                   &ndc->speed, &ndc->duplex);
1243 }
1244
1245 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
1246 {
1247         struct net_device_context *ndevctx = netdev_priv(ndev);
1248         struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1249         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1250         int orig_mtu = ndev->mtu;
1251         struct netvsc_device_info *device_info;
1252         int ret = 0;
1253
1254         if (!nvdev || nvdev->destroy)
1255                 return -ENODEV;
1256
1257         device_info = netvsc_devinfo_get(nvdev);
1258
1259         if (!device_info)
1260                 return -ENOMEM;
1261
1262         /* Change MTU of underlying VF netdev first. */
1263         if (vf_netdev) {
1264                 ret = dev_set_mtu(vf_netdev, mtu);
1265                 if (ret)
1266                         goto out;
1267         }
1268
1269         ret = netvsc_detach(ndev, nvdev);
1270         if (ret)
1271                 goto rollback_vf;
1272
1273         ndev->mtu = mtu;
1274
1275         ret = netvsc_attach(ndev, device_info);
1276         if (!ret)
1277                 goto out;
1278
1279         /* Attempt rollback to original MTU */
1280         ndev->mtu = orig_mtu;
1281
1282         if (netvsc_attach(ndev, device_info))
1283                 netdev_err(ndev, "restoring mtu failed\n");
1284 rollback_vf:
1285         if (vf_netdev)
1286                 dev_set_mtu(vf_netdev, orig_mtu);
1287
1288 out:
1289         netvsc_devinfo_put(device_info);
1290         return ret;
1291 }
1292
1293 static void netvsc_get_vf_stats(struct net_device *net,
1294                                 struct netvsc_vf_pcpu_stats *tot)
1295 {
1296         struct net_device_context *ndev_ctx = netdev_priv(net);
1297         int i;
1298
1299         memset(tot, 0, sizeof(*tot));
1300
1301         for_each_possible_cpu(i) {
1302                 const struct netvsc_vf_pcpu_stats *stats
1303                         = per_cpu_ptr(ndev_ctx->vf_stats, i);
1304                 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1305                 unsigned int start;
1306
1307                 do {
1308                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1309                         rx_packets = stats->rx_packets;
1310                         tx_packets = stats->tx_packets;
1311                         rx_bytes = stats->rx_bytes;
1312                         tx_bytes = stats->tx_bytes;
1313                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1314
1315                 tot->rx_packets += rx_packets;
1316                 tot->tx_packets += tx_packets;
1317                 tot->rx_bytes   += rx_bytes;
1318                 tot->tx_bytes   += tx_bytes;
1319                 tot->tx_dropped += stats->tx_dropped;
1320         }
1321 }
1322
1323 static void netvsc_get_pcpu_stats(struct net_device *net,
1324                                   struct netvsc_ethtool_pcpu_stats *pcpu_tot)
1325 {
1326         struct net_device_context *ndev_ctx = netdev_priv(net);
1327         struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
1328         int i;
1329
1330         /* fetch percpu stats of vf */
1331         for_each_possible_cpu(i) {
1332                 const struct netvsc_vf_pcpu_stats *stats =
1333                         per_cpu_ptr(ndev_ctx->vf_stats, i);
1334                 struct netvsc_ethtool_pcpu_stats *this_tot = &pcpu_tot[i];
1335                 unsigned int start;
1336
1337                 do {
1338                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1339                         this_tot->vf_rx_packets = stats->rx_packets;
1340                         this_tot->vf_tx_packets = stats->tx_packets;
1341                         this_tot->vf_rx_bytes = stats->rx_bytes;
1342                         this_tot->vf_tx_bytes = stats->tx_bytes;
1343                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1344                 this_tot->rx_packets = this_tot->vf_rx_packets;
1345                 this_tot->tx_packets = this_tot->vf_tx_packets;
1346                 this_tot->rx_bytes   = this_tot->vf_rx_bytes;
1347                 this_tot->tx_bytes   = this_tot->vf_tx_bytes;
1348         }
1349
1350         /* fetch percpu stats of netvsc */
1351         for (i = 0; i < nvdev->num_chn; i++) {
1352                 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1353                 const struct netvsc_stats *stats;
1354                 struct netvsc_ethtool_pcpu_stats *this_tot =
1355                         &pcpu_tot[nvchan->channel->target_cpu];
1356                 u64 packets, bytes;
1357                 unsigned int start;
1358
1359                 stats = &nvchan->tx_stats;
1360                 do {
1361                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1362                         packets = stats->packets;
1363                         bytes = stats->bytes;
1364                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1365
1366                 this_tot->tx_bytes      += bytes;
1367                 this_tot->tx_packets    += packets;
1368
1369                 stats = &nvchan->rx_stats;
1370                 do {
1371                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1372                         packets = stats->packets;
1373                         bytes = stats->bytes;
1374                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1375
1376                 this_tot->rx_bytes      += bytes;
1377                 this_tot->rx_packets    += packets;
1378         }
1379 }
1380
1381 static void netvsc_get_stats64(struct net_device *net,
1382                                struct rtnl_link_stats64 *t)
1383 {
1384         struct net_device_context *ndev_ctx = netdev_priv(net);
1385         struct netvsc_device *nvdev;
1386         struct netvsc_vf_pcpu_stats vf_tot;
1387         int i;
1388
1389         rcu_read_lock();
1390
1391         nvdev = rcu_dereference(ndev_ctx->nvdev);
1392         if (!nvdev)
1393                 goto out;
1394
1395         netdev_stats_to_stats64(t, &net->stats);
1396
1397         netvsc_get_vf_stats(net, &vf_tot);
1398         t->rx_packets += vf_tot.rx_packets;
1399         t->tx_packets += vf_tot.tx_packets;
1400         t->rx_bytes   += vf_tot.rx_bytes;
1401         t->tx_bytes   += vf_tot.tx_bytes;
1402         t->tx_dropped += vf_tot.tx_dropped;
1403
1404         for (i = 0; i < nvdev->num_chn; i++) {
1405                 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1406                 const struct netvsc_stats *stats;
1407                 u64 packets, bytes, multicast;
1408                 unsigned int start;
1409
1410                 stats = &nvchan->tx_stats;
1411                 do {
1412                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1413                         packets = stats->packets;
1414                         bytes = stats->bytes;
1415                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1416
1417                 t->tx_bytes     += bytes;
1418                 t->tx_packets   += packets;
1419
1420                 stats = &nvchan->rx_stats;
1421                 do {
1422                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1423                         packets = stats->packets;
1424                         bytes = stats->bytes;
1425                         multicast = stats->multicast + stats->broadcast;
1426                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1427
1428                 t->rx_bytes     += bytes;
1429                 t->rx_packets   += packets;
1430                 t->multicast    += multicast;
1431         }
1432 out:
1433         rcu_read_unlock();
1434 }
1435
1436 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
1437 {
1438         struct net_device_context *ndc = netdev_priv(ndev);
1439         struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1440         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1441         struct sockaddr *addr = p;
1442         int err;
1443
1444         err = eth_prepare_mac_addr_change(ndev, p);
1445         if (err)
1446                 return err;
1447
1448         if (!nvdev)
1449                 return -ENODEV;
1450
1451         if (vf_netdev) {
1452                 err = dev_set_mac_address(vf_netdev, addr, NULL);
1453                 if (err)
1454                         return err;
1455         }
1456
1457         err = rndis_filter_set_device_mac(nvdev, addr->sa_data);
1458         if (!err) {
1459                 eth_commit_mac_addr_change(ndev, p);
1460         } else if (vf_netdev) {
1461                 /* rollback change on VF */
1462                 memcpy(addr->sa_data, ndev->dev_addr, ETH_ALEN);
1463                 dev_set_mac_address(vf_netdev, addr, NULL);
1464         }
1465
1466         return err;
1467 }
1468
1469 static const struct {
1470         char name[ETH_GSTRING_LEN];
1471         u16 offset;
1472 } netvsc_stats[] = {
1473         { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
1474         { "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
1475         { "tx_no_space",  offsetof(struct netvsc_ethtool_stats, tx_no_space) },
1476         { "tx_too_big",   offsetof(struct netvsc_ethtool_stats, tx_too_big) },
1477         { "tx_busy",      offsetof(struct netvsc_ethtool_stats, tx_busy) },
1478         { "tx_send_full", offsetof(struct netvsc_ethtool_stats, tx_send_full) },
1479         { "rx_comp_busy", offsetof(struct netvsc_ethtool_stats, rx_comp_busy) },
1480         { "rx_no_memory", offsetof(struct netvsc_ethtool_stats, rx_no_memory) },
1481         { "stop_queue", offsetof(struct netvsc_ethtool_stats, stop_queue) },
1482         { "wake_queue", offsetof(struct netvsc_ethtool_stats, wake_queue) },
1483         { "vlan_error", offsetof(struct netvsc_ethtool_stats, vlan_error) },
1484 }, pcpu_stats[] = {
1485         { "cpu%u_rx_packets",
1486                 offsetof(struct netvsc_ethtool_pcpu_stats, rx_packets) },
1487         { "cpu%u_rx_bytes",
1488                 offsetof(struct netvsc_ethtool_pcpu_stats, rx_bytes) },
1489         { "cpu%u_tx_packets",
1490                 offsetof(struct netvsc_ethtool_pcpu_stats, tx_packets) },
1491         { "cpu%u_tx_bytes",
1492                 offsetof(struct netvsc_ethtool_pcpu_stats, tx_bytes) },
1493         { "cpu%u_vf_rx_packets",
1494                 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_packets) },
1495         { "cpu%u_vf_rx_bytes",
1496                 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_bytes) },
1497         { "cpu%u_vf_tx_packets",
1498                 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_packets) },
1499         { "cpu%u_vf_tx_bytes",
1500                 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_bytes) },
1501 }, vf_stats[] = {
1502         { "vf_rx_packets", offsetof(struct netvsc_vf_pcpu_stats, rx_packets) },
1503         { "vf_rx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, rx_bytes) },
1504         { "vf_tx_packets", offsetof(struct netvsc_vf_pcpu_stats, tx_packets) },
1505         { "vf_tx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, tx_bytes) },
1506         { "vf_tx_dropped", offsetof(struct netvsc_vf_pcpu_stats, tx_dropped) },
1507 };
1508
1509 #define NETVSC_GLOBAL_STATS_LEN ARRAY_SIZE(netvsc_stats)
1510 #define NETVSC_VF_STATS_LEN     ARRAY_SIZE(vf_stats)
1511
1512 /* statistics per queue (rx/tx packets/bytes) */
1513 #define NETVSC_PCPU_STATS_LEN (num_present_cpus() * ARRAY_SIZE(pcpu_stats))
1514
1515 /* 5 statistics per queue (rx/tx packets/bytes, rx xdp_drop) */
1516 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 5)
1517
1518 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
1519 {
1520         struct net_device_context *ndc = netdev_priv(dev);
1521         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1522
1523         if (!nvdev)
1524                 return -ENODEV;
1525
1526         switch (string_set) {
1527         case ETH_SS_STATS:
1528                 return NETVSC_GLOBAL_STATS_LEN
1529                         + NETVSC_VF_STATS_LEN
1530                         + NETVSC_QUEUE_STATS_LEN(nvdev)
1531                         + NETVSC_PCPU_STATS_LEN;
1532         default:
1533                 return -EINVAL;
1534         }
1535 }
1536
1537 static void netvsc_get_ethtool_stats(struct net_device *dev,
1538                                      struct ethtool_stats *stats, u64 *data)
1539 {
1540         struct net_device_context *ndc = netdev_priv(dev);
1541         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1542         const void *nds = &ndc->eth_stats;
1543         const struct netvsc_stats *qstats;
1544         struct netvsc_vf_pcpu_stats sum;
1545         struct netvsc_ethtool_pcpu_stats *pcpu_sum;
1546         unsigned int start;
1547         u64 packets, bytes;
1548         u64 xdp_drop;
1549         int i, j, cpu;
1550
1551         if (!nvdev)
1552                 return;
1553
1554         for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
1555                 data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1556
1557         netvsc_get_vf_stats(dev, &sum);
1558         for (j = 0; j < NETVSC_VF_STATS_LEN; j++)
1559                 data[i++] = *(u64 *)((void *)&sum + vf_stats[j].offset);
1560
1561         for (j = 0; j < nvdev->num_chn; j++) {
1562                 qstats = &nvdev->chan_table[j].tx_stats;
1563
1564                 do {
1565                         start = u64_stats_fetch_begin_irq(&qstats->syncp);
1566                         packets = qstats->packets;
1567                         bytes = qstats->bytes;
1568                 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1569                 data[i++] = packets;
1570                 data[i++] = bytes;
1571
1572                 qstats = &nvdev->chan_table[j].rx_stats;
1573                 do {
1574                         start = u64_stats_fetch_begin_irq(&qstats->syncp);
1575                         packets = qstats->packets;
1576                         bytes = qstats->bytes;
1577                         xdp_drop = qstats->xdp_drop;
1578                 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1579                 data[i++] = packets;
1580                 data[i++] = bytes;
1581                 data[i++] = xdp_drop;
1582         }
1583
1584         pcpu_sum = kvmalloc_array(num_possible_cpus(),
1585                                   sizeof(struct netvsc_ethtool_pcpu_stats),
1586                                   GFP_KERNEL);
1587         netvsc_get_pcpu_stats(dev, pcpu_sum);
1588         for_each_present_cpu(cpu) {
1589                 struct netvsc_ethtool_pcpu_stats *this_sum = &pcpu_sum[cpu];
1590
1591                 for (j = 0; j < ARRAY_SIZE(pcpu_stats); j++)
1592                         data[i++] = *(u64 *)((void *)this_sum
1593                                              + pcpu_stats[j].offset);
1594         }
1595         kvfree(pcpu_sum);
1596 }
1597
1598 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1599 {
1600         struct net_device_context *ndc = netdev_priv(dev);
1601         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1602         u8 *p = data;
1603         int i, cpu;
1604
1605         if (!nvdev)
1606                 return;
1607
1608         switch (stringset) {
1609         case ETH_SS_STATS:
1610                 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++) {
1611                         memcpy(p, netvsc_stats[i].name, ETH_GSTRING_LEN);
1612                         p += ETH_GSTRING_LEN;
1613                 }
1614
1615                 for (i = 0; i < ARRAY_SIZE(vf_stats); i++) {
1616                         memcpy(p, vf_stats[i].name, ETH_GSTRING_LEN);
1617                         p += ETH_GSTRING_LEN;
1618                 }
1619
1620                 for (i = 0; i < nvdev->num_chn; i++) {
1621                         sprintf(p, "tx_queue_%u_packets", i);
1622                         p += ETH_GSTRING_LEN;
1623                         sprintf(p, "tx_queue_%u_bytes", i);
1624                         p += ETH_GSTRING_LEN;
1625                         sprintf(p, "rx_queue_%u_packets", i);
1626                         p += ETH_GSTRING_LEN;
1627                         sprintf(p, "rx_queue_%u_bytes", i);
1628                         p += ETH_GSTRING_LEN;
1629                         sprintf(p, "rx_queue_%u_xdp_drop", i);
1630                         p += ETH_GSTRING_LEN;
1631                 }
1632
1633                 for_each_present_cpu(cpu) {
1634                         for (i = 0; i < ARRAY_SIZE(pcpu_stats); i++) {
1635                                 sprintf(p, pcpu_stats[i].name, cpu);
1636                                 p += ETH_GSTRING_LEN;
1637                         }
1638                 }
1639
1640                 break;
1641         }
1642 }
1643
1644 static int
1645 netvsc_get_rss_hash_opts(struct net_device_context *ndc,
1646                          struct ethtool_rxnfc *info)
1647 {
1648         const u32 l4_flag = RXH_L4_B_0_1 | RXH_L4_B_2_3;
1649
1650         info->data = RXH_IP_SRC | RXH_IP_DST;
1651
1652         switch (info->flow_type) {
1653         case TCP_V4_FLOW:
1654                 if (ndc->l4_hash & HV_TCP4_L4HASH)
1655                         info->data |= l4_flag;
1656
1657                 break;
1658
1659         case TCP_V6_FLOW:
1660                 if (ndc->l4_hash & HV_TCP6_L4HASH)
1661                         info->data |= l4_flag;
1662
1663                 break;
1664
1665         case UDP_V4_FLOW:
1666                 if (ndc->l4_hash & HV_UDP4_L4HASH)
1667                         info->data |= l4_flag;
1668
1669                 break;
1670
1671         case UDP_V6_FLOW:
1672                 if (ndc->l4_hash & HV_UDP6_L4HASH)
1673                         info->data |= l4_flag;
1674
1675                 break;
1676
1677         case IPV4_FLOW:
1678         case IPV6_FLOW:
1679                 break;
1680         default:
1681                 info->data = 0;
1682                 break;
1683         }
1684
1685         return 0;
1686 }
1687
1688 static int
1689 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1690                  u32 *rules)
1691 {
1692         struct net_device_context *ndc = netdev_priv(dev);
1693         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1694
1695         if (!nvdev)
1696                 return -ENODEV;
1697
1698         switch (info->cmd) {
1699         case ETHTOOL_GRXRINGS:
1700                 info->data = nvdev->num_chn;
1701                 return 0;
1702
1703         case ETHTOOL_GRXFH:
1704                 return netvsc_get_rss_hash_opts(ndc, info);
1705         }
1706         return -EOPNOTSUPP;
1707 }
1708
1709 static int netvsc_set_rss_hash_opts(struct net_device_context *ndc,
1710                                     struct ethtool_rxnfc *info)
1711 {
1712         if (info->data == (RXH_IP_SRC | RXH_IP_DST |
1713                            RXH_L4_B_0_1 | RXH_L4_B_2_3)) {
1714                 switch (info->flow_type) {
1715                 case TCP_V4_FLOW:
1716                         ndc->l4_hash |= HV_TCP4_L4HASH;
1717                         break;
1718
1719                 case TCP_V6_FLOW:
1720                         ndc->l4_hash |= HV_TCP6_L4HASH;
1721                         break;
1722
1723                 case UDP_V4_FLOW:
1724                         ndc->l4_hash |= HV_UDP4_L4HASH;
1725                         break;
1726
1727                 case UDP_V6_FLOW:
1728                         ndc->l4_hash |= HV_UDP6_L4HASH;
1729                         break;
1730
1731                 default:
1732                         return -EOPNOTSUPP;
1733                 }
1734
1735                 return 0;
1736         }
1737
1738         if (info->data == (RXH_IP_SRC | RXH_IP_DST)) {
1739                 switch (info->flow_type) {
1740                 case TCP_V4_FLOW:
1741                         ndc->l4_hash &= ~HV_TCP4_L4HASH;
1742                         break;
1743
1744                 case TCP_V6_FLOW:
1745                         ndc->l4_hash &= ~HV_TCP6_L4HASH;
1746                         break;
1747
1748                 case UDP_V4_FLOW:
1749                         ndc->l4_hash &= ~HV_UDP4_L4HASH;
1750                         break;
1751
1752                 case UDP_V6_FLOW:
1753                         ndc->l4_hash &= ~HV_UDP6_L4HASH;
1754                         break;
1755
1756                 default:
1757                         return -EOPNOTSUPP;
1758                 }
1759
1760                 return 0;
1761         }
1762
1763         return -EOPNOTSUPP;
1764 }
1765
1766 static int
1767 netvsc_set_rxnfc(struct net_device *ndev, struct ethtool_rxnfc *info)
1768 {
1769         struct net_device_context *ndc = netdev_priv(ndev);
1770
1771         if (info->cmd == ETHTOOL_SRXFH)
1772                 return netvsc_set_rss_hash_opts(ndc, info);
1773
1774         return -EOPNOTSUPP;
1775 }
1776
1777 static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1778 {
1779         return NETVSC_HASH_KEYLEN;
1780 }
1781
1782 static u32 netvsc_rss_indir_size(struct net_device *dev)
1783 {
1784         return ITAB_NUM;
1785 }
1786
1787 static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key,
1788                            u8 *hfunc)
1789 {
1790         struct net_device_context *ndc = netdev_priv(dev);
1791         struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1792         struct rndis_device *rndis_dev;
1793         int i;
1794
1795         if (!ndev)
1796                 return -ENODEV;
1797
1798         if (hfunc)
1799                 *hfunc = ETH_RSS_HASH_TOP;      /* Toeplitz */
1800
1801         rndis_dev = ndev->extension;
1802         if (indir) {
1803                 for (i = 0; i < ITAB_NUM; i++)
1804                         indir[i] = ndc->rx_table[i];
1805         }
1806
1807         if (key)
1808                 memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1809
1810         return 0;
1811 }
1812
1813 static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
1814                            const u8 *key, const u8 hfunc)
1815 {
1816         struct net_device_context *ndc = netdev_priv(dev);
1817         struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1818         struct rndis_device *rndis_dev;
1819         int i;
1820
1821         if (!ndev)
1822                 return -ENODEV;
1823
1824         if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
1825                 return -EOPNOTSUPP;
1826
1827         rndis_dev = ndev->extension;
1828         if (indir) {
1829                 for (i = 0; i < ITAB_NUM; i++)
1830                         if (indir[i] >= ndev->num_chn)
1831                                 return -EINVAL;
1832
1833                 for (i = 0; i < ITAB_NUM; i++)
1834                         ndc->rx_table[i] = indir[i];
1835         }
1836
1837         if (!key) {
1838                 if (!indir)
1839                         return 0;
1840
1841                 key = rndis_dev->rss_key;
1842         }
1843
1844         return rndis_filter_set_rss_param(rndis_dev, key);
1845 }
1846
1847 /* Hyper-V RNDIS protocol does not have ring in the HW sense.
1848  * It does have pre-allocated receive area which is divided into sections.
1849  */
1850 static void __netvsc_get_ringparam(struct netvsc_device *nvdev,
1851                                    struct ethtool_ringparam *ring)
1852 {
1853         u32 max_buf_size;
1854
1855         ring->rx_pending = nvdev->recv_section_cnt;
1856         ring->tx_pending = nvdev->send_section_cnt;
1857
1858         if (nvdev->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
1859                 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
1860         else
1861                 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
1862
1863         ring->rx_max_pending = max_buf_size / nvdev->recv_section_size;
1864         ring->tx_max_pending = NETVSC_SEND_BUFFER_SIZE
1865                 / nvdev->send_section_size;
1866 }
1867
1868 static void netvsc_get_ringparam(struct net_device *ndev,
1869                                  struct ethtool_ringparam *ring)
1870 {
1871         struct net_device_context *ndevctx = netdev_priv(ndev);
1872         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1873
1874         if (!nvdev)
1875                 return;
1876
1877         __netvsc_get_ringparam(nvdev, ring);
1878 }
1879
1880 static int netvsc_set_ringparam(struct net_device *ndev,
1881                                 struct ethtool_ringparam *ring)
1882 {
1883         struct net_device_context *ndevctx = netdev_priv(ndev);
1884         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1885         struct netvsc_device_info *device_info;
1886         struct ethtool_ringparam orig;
1887         u32 new_tx, new_rx;
1888         int ret = 0;
1889
1890         if (!nvdev || nvdev->destroy)
1891                 return -ENODEV;
1892
1893         memset(&orig, 0, sizeof(orig));
1894         __netvsc_get_ringparam(nvdev, &orig);
1895
1896         new_tx = clamp_t(u32, ring->tx_pending,
1897                          NETVSC_MIN_TX_SECTIONS, orig.tx_max_pending);
1898         new_rx = clamp_t(u32, ring->rx_pending,
1899                          NETVSC_MIN_RX_SECTIONS, orig.rx_max_pending);
1900
1901         if (new_tx == orig.tx_pending &&
1902             new_rx == orig.rx_pending)
1903                 return 0;        /* no change */
1904
1905         device_info = netvsc_devinfo_get(nvdev);
1906
1907         if (!device_info)
1908                 return -ENOMEM;
1909
1910         device_info->send_sections = new_tx;
1911         device_info->recv_sections = new_rx;
1912
1913         ret = netvsc_detach(ndev, nvdev);
1914         if (ret)
1915                 goto out;
1916
1917         ret = netvsc_attach(ndev, device_info);
1918         if (ret) {
1919                 device_info->send_sections = orig.tx_pending;
1920                 device_info->recv_sections = orig.rx_pending;
1921
1922                 if (netvsc_attach(ndev, device_info))
1923                         netdev_err(ndev, "restoring ringparam failed");
1924         }
1925
1926 out:
1927         netvsc_devinfo_put(device_info);
1928         return ret;
1929 }
1930
1931 static netdev_features_t netvsc_fix_features(struct net_device *ndev,
1932                                              netdev_features_t features)
1933 {
1934         struct net_device_context *ndevctx = netdev_priv(ndev);
1935         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1936
1937         if (!nvdev || nvdev->destroy)
1938                 return features;
1939
1940         if ((features & NETIF_F_LRO) && netvsc_xdp_get(nvdev)) {
1941                 features ^= NETIF_F_LRO;
1942                 netdev_info(ndev, "Skip LRO - unsupported with XDP\n");
1943         }
1944
1945         return features;
1946 }
1947
1948 static int netvsc_set_features(struct net_device *ndev,
1949                                netdev_features_t features)
1950 {
1951         netdev_features_t change = features ^ ndev->features;
1952         struct net_device_context *ndevctx = netdev_priv(ndev);
1953         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1954         struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1955         struct ndis_offload_params offloads;
1956         int ret = 0;
1957
1958         if (!nvdev || nvdev->destroy)
1959                 return -ENODEV;
1960
1961         if (!(change & NETIF_F_LRO))
1962                 goto syncvf;
1963
1964         memset(&offloads, 0, sizeof(struct ndis_offload_params));
1965
1966         if (features & NETIF_F_LRO) {
1967                 offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1968                 offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1969         } else {
1970                 offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1971                 offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1972         }
1973
1974         ret = rndis_filter_set_offload_params(ndev, nvdev, &offloads);
1975
1976         if (ret) {
1977                 features ^= NETIF_F_LRO;
1978                 ndev->features = features;
1979         }
1980
1981 syncvf:
1982         if (!vf_netdev)
1983                 return ret;
1984
1985         vf_netdev->wanted_features = features;
1986         netdev_update_features(vf_netdev);
1987
1988         return ret;
1989 }
1990
1991 static int netvsc_get_regs_len(struct net_device *netdev)
1992 {
1993         return VRSS_SEND_TAB_SIZE * sizeof(u32);
1994 }
1995
1996 static void netvsc_get_regs(struct net_device *netdev,
1997                             struct ethtool_regs *regs, void *p)
1998 {
1999         struct net_device_context *ndc = netdev_priv(netdev);
2000         u32 *regs_buff = p;
2001
2002         /* increase the version, if buffer format is changed. */
2003         regs->version = 1;
2004
2005         memcpy(regs_buff, ndc->tx_table, VRSS_SEND_TAB_SIZE * sizeof(u32));
2006 }
2007
2008 static u32 netvsc_get_msglevel(struct net_device *ndev)
2009 {
2010         struct net_device_context *ndev_ctx = netdev_priv(ndev);
2011
2012         return ndev_ctx->msg_enable;
2013 }
2014
2015 static void netvsc_set_msglevel(struct net_device *ndev, u32 val)
2016 {
2017         struct net_device_context *ndev_ctx = netdev_priv(ndev);
2018
2019         ndev_ctx->msg_enable = val;
2020 }
2021
2022 static const struct ethtool_ops ethtool_ops = {
2023         .get_drvinfo    = netvsc_get_drvinfo,
2024         .get_regs_len   = netvsc_get_regs_len,
2025         .get_regs       = netvsc_get_regs,
2026         .get_msglevel   = netvsc_get_msglevel,
2027         .set_msglevel   = netvsc_set_msglevel,
2028         .get_link       = ethtool_op_get_link,
2029         .get_ethtool_stats = netvsc_get_ethtool_stats,
2030         .get_sset_count = netvsc_get_sset_count,
2031         .get_strings    = netvsc_get_strings,
2032         .get_channels   = netvsc_get_channels,
2033         .set_channels   = netvsc_set_channels,
2034         .get_ts_info    = ethtool_op_get_ts_info,
2035         .get_rxnfc      = netvsc_get_rxnfc,
2036         .set_rxnfc      = netvsc_set_rxnfc,
2037         .get_rxfh_key_size = netvsc_get_rxfh_key_size,
2038         .get_rxfh_indir_size = netvsc_rss_indir_size,
2039         .get_rxfh       = netvsc_get_rxfh,
2040         .set_rxfh       = netvsc_set_rxfh,
2041         .get_link_ksettings = netvsc_get_link_ksettings,
2042         .set_link_ksettings = netvsc_set_link_ksettings,
2043         .get_ringparam  = netvsc_get_ringparam,
2044         .set_ringparam  = netvsc_set_ringparam,
2045 };
2046
2047 static const struct net_device_ops device_ops = {
2048         .ndo_open =                     netvsc_open,
2049         .ndo_stop =                     netvsc_close,
2050         .ndo_start_xmit =               netvsc_start_xmit,
2051         .ndo_change_rx_flags =          netvsc_change_rx_flags,
2052         .ndo_set_rx_mode =              netvsc_set_rx_mode,
2053         .ndo_fix_features =             netvsc_fix_features,
2054         .ndo_set_features =             netvsc_set_features,
2055         .ndo_change_mtu =               netvsc_change_mtu,
2056         .ndo_validate_addr =            eth_validate_addr,
2057         .ndo_set_mac_address =          netvsc_set_mac_addr,
2058         .ndo_select_queue =             netvsc_select_queue,
2059         .ndo_get_stats64 =              netvsc_get_stats64,
2060         .ndo_bpf =                      netvsc_bpf,
2061 };
2062
2063 /*
2064  * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
2065  * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
2066  * present send GARP packet to network peers with netif_notify_peers().
2067  */
2068 static void netvsc_link_change(struct work_struct *w)
2069 {
2070         struct net_device_context *ndev_ctx =
2071                 container_of(w, struct net_device_context, dwork.work);
2072         struct hv_device *device_obj = ndev_ctx->device_ctx;
2073         struct net_device *net = hv_get_drvdata(device_obj);
2074         unsigned long flags, next_reconfig, delay;
2075         struct netvsc_reconfig *event = NULL;
2076         struct netvsc_device *net_device;
2077         struct rndis_device *rdev;
2078         bool reschedule = false;
2079
2080         /* if changes are happening, comeback later */
2081         if (!rtnl_trylock()) {
2082                 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
2083                 return;
2084         }
2085
2086         net_device = rtnl_dereference(ndev_ctx->nvdev);
2087         if (!net_device)
2088                 goto out_unlock;
2089
2090         rdev = net_device->extension;
2091
2092         next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
2093         if (time_is_after_jiffies(next_reconfig)) {
2094                 /* link_watch only sends one notification with current state
2095                  * per second, avoid doing reconfig more frequently. Handle
2096                  * wrap around.
2097                  */
2098                 delay = next_reconfig - jiffies;
2099                 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
2100                 schedule_delayed_work(&ndev_ctx->dwork, delay);
2101                 goto out_unlock;
2102         }
2103         ndev_ctx->last_reconfig = jiffies;
2104
2105         spin_lock_irqsave(&ndev_ctx->lock, flags);
2106         if (!list_empty(&ndev_ctx->reconfig_events)) {
2107                 event = list_first_entry(&ndev_ctx->reconfig_events,
2108                                          struct netvsc_reconfig, list);
2109                 list_del(&event->list);
2110                 reschedule = !list_empty(&ndev_ctx->reconfig_events);
2111         }
2112         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
2113
2114         if (!event)
2115                 goto out_unlock;
2116
2117         switch (event->event) {
2118                 /* Only the following events are possible due to the check in
2119                  * netvsc_linkstatus_callback()
2120                  */
2121         case RNDIS_STATUS_MEDIA_CONNECT:
2122                 if (rdev->link_state) {
2123                         rdev->link_state = false;
2124                         netif_carrier_on(net);
2125                         netvsc_tx_enable(net_device, net);
2126                 } else {
2127                         __netdev_notify_peers(net);
2128                 }
2129                 kfree(event);
2130                 break;
2131         case RNDIS_STATUS_MEDIA_DISCONNECT:
2132                 if (!rdev->link_state) {
2133                         rdev->link_state = true;
2134                         netif_carrier_off(net);
2135                         netvsc_tx_disable(net_device, net);
2136                 }
2137                 kfree(event);
2138                 break;
2139         case RNDIS_STATUS_NETWORK_CHANGE:
2140                 /* Only makes sense if carrier is present */
2141                 if (!rdev->link_state) {
2142                         rdev->link_state = true;
2143                         netif_carrier_off(net);
2144                         netvsc_tx_disable(net_device, net);
2145                         event->event = RNDIS_STATUS_MEDIA_CONNECT;
2146                         spin_lock_irqsave(&ndev_ctx->lock, flags);
2147                         list_add(&event->list, &ndev_ctx->reconfig_events);
2148                         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
2149                         reschedule = true;
2150                 }
2151                 break;
2152         }
2153
2154         rtnl_unlock();
2155
2156         /* link_watch only sends one notification with current state per
2157          * second, handle next reconfig event in 2 seconds.
2158          */
2159         if (reschedule)
2160                 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
2161
2162         return;
2163
2164 out_unlock:
2165         rtnl_unlock();
2166 }
2167
2168 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
2169 {
2170         struct net_device_context *net_device_ctx;
2171         struct net_device *dev;
2172
2173         dev = netdev_master_upper_dev_get(vf_netdev);
2174         if (!dev || dev->netdev_ops != &device_ops)
2175                 return NULL;    /* not a netvsc device */
2176
2177         net_device_ctx = netdev_priv(dev);
2178         if (!rtnl_dereference(net_device_ctx->nvdev))
2179                 return NULL;    /* device is removed */
2180
2181         return dev;
2182 }
2183
2184 /* Called when VF is injecting data into network stack.
2185  * Change the associated network device from VF to netvsc.
2186  * note: already called with rcu_read_lock
2187  */
2188 static rx_handler_result_t netvsc_vf_handle_frame(struct sk_buff **pskb)
2189 {
2190         struct sk_buff *skb = *pskb;
2191         struct net_device *ndev = rcu_dereference(skb->dev->rx_handler_data);
2192         struct net_device_context *ndev_ctx = netdev_priv(ndev);
2193         struct netvsc_vf_pcpu_stats *pcpu_stats
2194                  = this_cpu_ptr(ndev_ctx->vf_stats);
2195
2196         skb = skb_share_check(skb, GFP_ATOMIC);
2197         if (unlikely(!skb))
2198                 return RX_HANDLER_CONSUMED;
2199
2200         *pskb = skb;
2201
2202         skb->dev = ndev;
2203
2204         u64_stats_update_begin(&pcpu_stats->syncp);
2205         pcpu_stats->rx_packets++;
2206         pcpu_stats->rx_bytes += skb->len;
2207         u64_stats_update_end(&pcpu_stats->syncp);
2208
2209         return RX_HANDLER_ANOTHER;
2210 }
2211
2212 static int netvsc_vf_join(struct net_device *vf_netdev,
2213                           struct net_device *ndev)
2214 {
2215         struct net_device_context *ndev_ctx = netdev_priv(ndev);
2216         int ret;
2217
2218         ret = netdev_rx_handler_register(vf_netdev,
2219                                          netvsc_vf_handle_frame, ndev);
2220         if (ret != 0) {
2221                 netdev_err(vf_netdev,
2222                            "can not register netvsc VF receive handler (err = %d)\n",
2223                            ret);
2224                 goto rx_handler_failed;
2225         }
2226
2227         ret = netdev_master_upper_dev_link(vf_netdev, ndev,
2228                                            NULL, NULL, NULL);
2229         if (ret != 0) {
2230                 netdev_err(vf_netdev,
2231                            "can not set master device %s (err = %d)\n",
2232                            ndev->name, ret);
2233                 goto upper_link_failed;
2234         }
2235
2236         /* set slave flag before open to prevent IPv6 addrconf */
2237         vf_netdev->flags |= IFF_SLAVE;
2238
2239         schedule_delayed_work(&ndev_ctx->vf_takeover, VF_TAKEOVER_INT);
2240
2241         call_netdevice_notifiers(NETDEV_JOIN, vf_netdev);
2242
2243         netdev_info(vf_netdev, "joined to %s\n", ndev->name);
2244         return 0;
2245
2246 upper_link_failed:
2247         netdev_rx_handler_unregister(vf_netdev);
2248 rx_handler_failed:
2249         return ret;
2250 }
2251
2252 static void __netvsc_vf_setup(struct net_device *ndev,
2253                               struct net_device *vf_netdev)
2254 {
2255         int ret;
2256
2257         /* Align MTU of VF with master */
2258         ret = dev_set_mtu(vf_netdev, ndev->mtu);
2259         if (ret)
2260                 netdev_warn(vf_netdev,
2261                             "unable to change mtu to %u\n", ndev->mtu);
2262
2263         /* set multicast etc flags on VF */
2264         dev_change_flags(vf_netdev, ndev->flags | IFF_SLAVE, NULL);
2265
2266         /* sync address list from ndev to VF */
2267         netif_addr_lock_bh(ndev);
2268         dev_uc_sync(vf_netdev, ndev);
2269         dev_mc_sync(vf_netdev, ndev);
2270         netif_addr_unlock_bh(ndev);
2271
2272         if (netif_running(ndev)) {
2273                 ret = dev_open(vf_netdev, NULL);
2274                 if (ret)
2275                         netdev_warn(vf_netdev,
2276                                     "unable to open: %d\n", ret);
2277         }
2278 }
2279
2280 /* Setup VF as slave of the synthetic device.
2281  * Runs in workqueue to avoid recursion in netlink callbacks.
2282  */
2283 static void netvsc_vf_setup(struct work_struct *w)
2284 {
2285         struct net_device_context *ndev_ctx
2286                 = container_of(w, struct net_device_context, vf_takeover.work);
2287         struct net_device *ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2288         struct net_device *vf_netdev;
2289
2290         if (!rtnl_trylock()) {
2291                 schedule_delayed_work(&ndev_ctx->vf_takeover, 0);
2292                 return;
2293         }
2294
2295         vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2296         if (vf_netdev)
2297                 __netvsc_vf_setup(ndev, vf_netdev);
2298
2299         rtnl_unlock();
2300 }
2301
2302 /* Find netvsc by VF serial number.
2303  * The PCI hyperv controller records the serial number as the slot kobj name.
2304  */
2305 static struct net_device *get_netvsc_byslot(const struct net_device *vf_netdev)
2306 {
2307         struct device *parent = vf_netdev->dev.parent;
2308         struct net_device_context *ndev_ctx;
2309         struct pci_dev *pdev;
2310         u32 serial;
2311
2312         if (!parent || !dev_is_pci(parent))
2313                 return NULL; /* not a PCI device */
2314
2315         pdev = to_pci_dev(parent);
2316         if (!pdev->slot) {
2317                 netdev_notice(vf_netdev, "no PCI slot information\n");
2318                 return NULL;
2319         }
2320
2321         if (kstrtou32(pci_slot_name(pdev->slot), 10, &serial)) {
2322                 netdev_notice(vf_netdev, "Invalid vf serial:%s\n",
2323                               pci_slot_name(pdev->slot));
2324                 return NULL;
2325         }
2326
2327         list_for_each_entry(ndev_ctx, &netvsc_dev_list, list) {
2328                 if (!ndev_ctx->vf_alloc)
2329                         continue;
2330
2331                 if (ndev_ctx->vf_serial == serial)
2332                         return hv_get_drvdata(ndev_ctx->device_ctx);
2333         }
2334
2335         netdev_notice(vf_netdev,
2336                       "no netdev found for vf serial:%u\n", serial);
2337         return NULL;
2338 }
2339
2340 static int netvsc_register_vf(struct net_device *vf_netdev)
2341 {
2342         struct net_device_context *net_device_ctx;
2343         struct netvsc_device *netvsc_dev;
2344         struct bpf_prog *prog;
2345         struct net_device *ndev;
2346         int ret;
2347
2348         if (vf_netdev->addr_len != ETH_ALEN)
2349                 return NOTIFY_DONE;
2350
2351         ndev = get_netvsc_byslot(vf_netdev);
2352         if (!ndev)
2353                 return NOTIFY_DONE;
2354
2355         net_device_ctx = netdev_priv(ndev);
2356         netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2357         if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
2358                 return NOTIFY_DONE;
2359
2360         /* if synthetic interface is a different namespace,
2361          * then move the VF to that namespace; join will be
2362          * done again in that context.
2363          */
2364         if (!net_eq(dev_net(ndev), dev_net(vf_netdev))) {
2365                 ret = dev_change_net_namespace(vf_netdev,
2366                                                dev_net(ndev), "eth%d");
2367                 if (ret)
2368                         netdev_err(vf_netdev,
2369                                    "could not move to same namespace as %s: %d\n",
2370                                    ndev->name, ret);
2371                 else
2372                         netdev_info(vf_netdev,
2373                                     "VF moved to namespace with: %s\n",
2374                                     ndev->name);
2375                 return NOTIFY_DONE;
2376         }
2377
2378         netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
2379
2380         if (netvsc_vf_join(vf_netdev, ndev) != 0)
2381                 return NOTIFY_DONE;
2382
2383         dev_hold(vf_netdev);
2384         rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
2385
2386         vf_netdev->wanted_features = ndev->features;
2387         netdev_update_features(vf_netdev);
2388
2389         prog = netvsc_xdp_get(netvsc_dev);
2390         netvsc_vf_setxdp(vf_netdev, prog);
2391
2392         return NOTIFY_OK;
2393 }
2394
2395 /* Change the data path when VF UP/DOWN/CHANGE are detected.
2396  *
2397  * Typically a UP or DOWN event is followed by a CHANGE event, so
2398  * net_device_ctx->data_path_is_vf is used to cache the current data path
2399  * to avoid the duplicate call of netvsc_switch_datapath() and the duplicate
2400  * message.
2401  *
2402  * During hibernation, if a VF NIC driver (e.g. mlx5) preserves the network
2403  * interface, there is only the CHANGE event and no UP or DOWN event.
2404  */
2405 static int netvsc_vf_changed(struct net_device *vf_netdev, unsigned long event)
2406 {
2407         struct net_device_context *net_device_ctx;
2408         struct netvsc_device *netvsc_dev;
2409         struct net_device *ndev;
2410         bool vf_is_up = false;
2411
2412         if (event != NETDEV_GOING_DOWN)
2413                 vf_is_up = netif_running(vf_netdev);
2414
2415         ndev = get_netvsc_byref(vf_netdev);
2416         if (!ndev)
2417                 return NOTIFY_DONE;
2418
2419         net_device_ctx = netdev_priv(ndev);
2420         netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2421         if (!netvsc_dev)
2422                 return NOTIFY_DONE;
2423
2424         if (net_device_ctx->data_path_is_vf == vf_is_up)
2425                 return NOTIFY_OK;
2426
2427         netvsc_switch_datapath(ndev, vf_is_up);
2428         netdev_info(ndev, "Data path switched %s VF: %s\n",
2429                     vf_is_up ? "to" : "from", vf_netdev->name);
2430
2431         return NOTIFY_OK;
2432 }
2433
2434 static int netvsc_unregister_vf(struct net_device *vf_netdev)
2435 {
2436         struct net_device *ndev;
2437         struct net_device_context *net_device_ctx;
2438
2439         ndev = get_netvsc_byref(vf_netdev);
2440         if (!ndev)
2441                 return NOTIFY_DONE;
2442
2443         net_device_ctx = netdev_priv(ndev);
2444         cancel_delayed_work_sync(&net_device_ctx->vf_takeover);
2445
2446         netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
2447
2448         netvsc_vf_setxdp(vf_netdev, NULL);
2449
2450         netdev_rx_handler_unregister(vf_netdev);
2451         netdev_upper_dev_unlink(vf_netdev, ndev);
2452         RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
2453         dev_put(vf_netdev);
2454
2455         return NOTIFY_OK;
2456 }
2457
2458 static int netvsc_probe(struct hv_device *dev,
2459                         const struct hv_vmbus_device_id *dev_id)
2460 {
2461         struct net_device *net = NULL;
2462         struct net_device_context *net_device_ctx;
2463         struct netvsc_device_info *device_info = NULL;
2464         struct netvsc_device *nvdev;
2465         int ret = -ENOMEM;
2466
2467         net = alloc_etherdev_mq(sizeof(struct net_device_context),
2468                                 VRSS_CHANNEL_MAX);
2469         if (!net)
2470                 goto no_net;
2471
2472         netif_carrier_off(net);
2473
2474         netvsc_init_settings(net);
2475
2476         net_device_ctx = netdev_priv(net);
2477         net_device_ctx->device_ctx = dev;
2478         net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
2479         if (netif_msg_probe(net_device_ctx))
2480                 netdev_dbg(net, "netvsc msg_enable: %d\n",
2481                            net_device_ctx->msg_enable);
2482
2483         hv_set_drvdata(dev, net);
2484
2485         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
2486
2487         spin_lock_init(&net_device_ctx->lock);
2488         INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
2489         INIT_DELAYED_WORK(&net_device_ctx->vf_takeover, netvsc_vf_setup);
2490
2491         net_device_ctx->vf_stats
2492                 = netdev_alloc_pcpu_stats(struct netvsc_vf_pcpu_stats);
2493         if (!net_device_ctx->vf_stats)
2494                 goto no_stats;
2495
2496         net->netdev_ops = &device_ops;
2497         net->ethtool_ops = &ethtool_ops;
2498         SET_NETDEV_DEV(net, &dev->device);
2499
2500         /* We always need headroom for rndis header */
2501         net->needed_headroom = RNDIS_AND_PPI_SIZE;
2502
2503         /* Initialize the number of queues to be 1, we may change it if more
2504          * channels are offered later.
2505          */
2506         netif_set_real_num_tx_queues(net, 1);
2507         netif_set_real_num_rx_queues(net, 1);
2508
2509         /* Notify the netvsc driver of the new device */
2510         device_info = netvsc_devinfo_get(NULL);
2511
2512         if (!device_info) {
2513                 ret = -ENOMEM;
2514                 goto devinfo_failed;
2515         }
2516
2517         nvdev = rndis_filter_device_add(dev, device_info);
2518         if (IS_ERR(nvdev)) {
2519                 ret = PTR_ERR(nvdev);
2520                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
2521                 goto rndis_failed;
2522         }
2523
2524         memcpy(net->dev_addr, device_info->mac_adr, ETH_ALEN);
2525
2526         /* We must get rtnl lock before scheduling nvdev->subchan_work,
2527          * otherwise netvsc_subchan_work() can get rtnl lock first and wait
2528          * all subchannels to show up, but that may not happen because
2529          * netvsc_probe() can't get rtnl lock and as a result vmbus_onoffer()
2530          * -> ... -> device_add() -> ... -> __device_attach() can't get
2531          * the device lock, so all the subchannels can't be processed --
2532          * finally netvsc_subchan_work() hangs forever.
2533          */
2534         rtnl_lock();
2535
2536         if (nvdev->num_chn > 1)
2537                 schedule_work(&nvdev->subchan_work);
2538
2539         /* hw_features computed in rndis_netdev_set_hwcaps() */
2540         net->features = net->hw_features |
2541                 NETIF_F_HIGHDMA | NETIF_F_HW_VLAN_CTAG_TX |
2542                 NETIF_F_HW_VLAN_CTAG_RX;
2543         net->vlan_features = net->features;
2544
2545         netdev_lockdep_set_classes(net);
2546
2547         /* MTU range: 68 - 1500 or 65521 */
2548         net->min_mtu = NETVSC_MTU_MIN;
2549         if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
2550                 net->max_mtu = NETVSC_MTU - ETH_HLEN;
2551         else
2552                 net->max_mtu = ETH_DATA_LEN;
2553
2554         nvdev->tx_disable = false;
2555
2556         ret = register_netdevice(net);
2557         if (ret != 0) {
2558                 pr_err("Unable to register netdev.\n");
2559                 goto register_failed;
2560         }
2561
2562         list_add(&net_device_ctx->list, &netvsc_dev_list);
2563         rtnl_unlock();
2564
2565         netvsc_devinfo_put(device_info);
2566         return 0;
2567
2568 register_failed:
2569         rtnl_unlock();
2570         rndis_filter_device_remove(dev, nvdev);
2571 rndis_failed:
2572         netvsc_devinfo_put(device_info);
2573 devinfo_failed:
2574         free_percpu(net_device_ctx->vf_stats);
2575 no_stats:
2576         hv_set_drvdata(dev, NULL);
2577         free_netdev(net);
2578 no_net:
2579         return ret;
2580 }
2581
2582 static int netvsc_remove(struct hv_device *dev)
2583 {
2584         struct net_device_context *ndev_ctx;
2585         struct net_device *vf_netdev, *net;
2586         struct netvsc_device *nvdev;
2587
2588         net = hv_get_drvdata(dev);
2589         if (net == NULL) {
2590                 dev_err(&dev->device, "No net device to remove\n");
2591                 return 0;
2592         }
2593
2594         ndev_ctx = netdev_priv(net);
2595
2596         cancel_delayed_work_sync(&ndev_ctx->dwork);
2597
2598         rtnl_lock();
2599         nvdev = rtnl_dereference(ndev_ctx->nvdev);
2600         if (nvdev) {
2601                 cancel_work_sync(&nvdev->subchan_work);
2602                 netvsc_xdp_set(net, NULL, NULL, nvdev);
2603         }
2604
2605         /*
2606          * Call to the vsc driver to let it know that the device is being
2607          * removed. Also blocks mtu and channel changes.
2608          */
2609         vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2610         if (vf_netdev)
2611                 netvsc_unregister_vf(vf_netdev);
2612
2613         if (nvdev)
2614                 rndis_filter_device_remove(dev, nvdev);
2615
2616         unregister_netdevice(net);
2617         list_del(&ndev_ctx->list);
2618
2619         rtnl_unlock();
2620
2621         hv_set_drvdata(dev, NULL);
2622
2623         free_percpu(ndev_ctx->vf_stats);
2624         free_netdev(net);
2625         return 0;
2626 }
2627
2628 static int netvsc_suspend(struct hv_device *dev)
2629 {
2630         struct net_device_context *ndev_ctx;
2631         struct netvsc_device *nvdev;
2632         struct net_device *net;
2633         int ret;
2634
2635         net = hv_get_drvdata(dev);
2636
2637         ndev_ctx = netdev_priv(net);
2638         cancel_delayed_work_sync(&ndev_ctx->dwork);
2639
2640         rtnl_lock();
2641
2642         nvdev = rtnl_dereference(ndev_ctx->nvdev);
2643         if (nvdev == NULL) {
2644                 ret = -ENODEV;
2645                 goto out;
2646         }
2647
2648         /* Save the current config info */
2649         ndev_ctx->saved_netvsc_dev_info = netvsc_devinfo_get(nvdev);
2650
2651         ret = netvsc_detach(net, nvdev);
2652 out:
2653         rtnl_unlock();
2654
2655         return ret;
2656 }
2657
2658 static int netvsc_resume(struct hv_device *dev)
2659 {
2660         struct net_device *net = hv_get_drvdata(dev);
2661         struct net_device_context *net_device_ctx;
2662         struct netvsc_device_info *device_info;
2663         int ret;
2664
2665         rtnl_lock();
2666
2667         net_device_ctx = netdev_priv(net);
2668
2669         /* Reset the data path to the netvsc NIC before re-opening the vmbus
2670          * channel. Later netvsc_netdev_event() will switch the data path to
2671          * the VF upon the UP or CHANGE event.
2672          */
2673         net_device_ctx->data_path_is_vf = false;
2674         device_info = net_device_ctx->saved_netvsc_dev_info;
2675
2676         ret = netvsc_attach(net, device_info);
2677
2678         netvsc_devinfo_put(device_info);
2679         net_device_ctx->saved_netvsc_dev_info = NULL;
2680
2681         rtnl_unlock();
2682
2683         return ret;
2684 }
2685 static const struct hv_vmbus_device_id id_table[] = {
2686         /* Network guid */
2687         { HV_NIC_GUID, },
2688         { },
2689 };
2690
2691 MODULE_DEVICE_TABLE(vmbus, id_table);
2692
2693 /* The one and only one */
2694 static struct  hv_driver netvsc_drv = {
2695         .name = KBUILD_MODNAME,
2696         .id_table = id_table,
2697         .probe = netvsc_probe,
2698         .remove = netvsc_remove,
2699         .suspend = netvsc_suspend,
2700         .resume = netvsc_resume,
2701         .driver = {
2702                 .probe_type = PROBE_FORCE_SYNCHRONOUS,
2703         },
2704 };
2705
2706 /*
2707  * On Hyper-V, every VF interface is matched with a corresponding
2708  * synthetic interface. The synthetic interface is presented first
2709  * to the guest. When the corresponding VF instance is registered,
2710  * we will take care of switching the data path.
2711  */
2712 static int netvsc_netdev_event(struct notifier_block *this,
2713                                unsigned long event, void *ptr)
2714 {
2715         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2716
2717         /* Skip our own events */
2718         if (event_dev->netdev_ops == &device_ops)
2719                 return NOTIFY_DONE;
2720
2721         /* Avoid non-Ethernet type devices */
2722         if (event_dev->type != ARPHRD_ETHER)
2723                 return NOTIFY_DONE;
2724
2725         /* Avoid Vlan dev with same MAC registering as VF */
2726         if (is_vlan_dev(event_dev))
2727                 return NOTIFY_DONE;
2728
2729         /* Avoid Bonding master dev with same MAC registering as VF */
2730         if ((event_dev->priv_flags & IFF_BONDING) &&
2731             (event_dev->flags & IFF_MASTER))
2732                 return NOTIFY_DONE;
2733
2734         switch (event) {
2735         case NETDEV_REGISTER:
2736                 return netvsc_register_vf(event_dev);
2737         case NETDEV_UNREGISTER:
2738                 return netvsc_unregister_vf(event_dev);
2739         case NETDEV_UP:
2740         case NETDEV_DOWN:
2741         case NETDEV_CHANGE:
2742         case NETDEV_GOING_DOWN:
2743                 return netvsc_vf_changed(event_dev, event);
2744         default:
2745                 return NOTIFY_DONE;
2746         }
2747 }
2748
2749 static struct notifier_block netvsc_netdev_notifier = {
2750         .notifier_call = netvsc_netdev_event,
2751 };
2752
2753 static void __exit netvsc_drv_exit(void)
2754 {
2755         unregister_netdevice_notifier(&netvsc_netdev_notifier);
2756         vmbus_driver_unregister(&netvsc_drv);
2757 }
2758
2759 static int __init netvsc_drv_init(void)
2760 {
2761         int ret;
2762
2763         if (ring_size < RING_SIZE_MIN) {
2764                 ring_size = RING_SIZE_MIN;
2765                 pr_info("Increased ring_size to %u (min allowed)\n",
2766                         ring_size);
2767         }
2768         netvsc_ring_bytes = ring_size * PAGE_SIZE;
2769
2770         ret = vmbus_driver_register(&netvsc_drv);
2771         if (ret)
2772                 return ret;
2773
2774         register_netdevice_notifier(&netvsc_netdev_notifier);
2775         return 0;
2776 }
2777
2778 MODULE_LICENSE("GPL");
2779 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
2780
2781 module_init(netvsc_drv_init);
2782 module_exit(netvsc_drv_exit);