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