drivers/net/hyperv/netvsc_drv.c

   1 /*
   2  * Copyright (c) 2009, Microsoft Corporation.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms and conditions of the GNU General Public License,
   6  * version 2, as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope it will be useful, but WITHOUT
   9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  11  * more details.
  12  *
  13  * You should have received a copy of the GNU General Public License along with
  14  * this program; if not, see <http://www.gnu.org/licenses/>.
  15  *
  16  * Authors:
  17  *   Haiyang Zhang <[email protected]>
  18  *   Hank Janssen  <[email protected]>
  19  */
  20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  21
  22 #include <linux/init.h>
  23 #include <linux/atomic.h>
  24 #include <linux/module.h>
  25 #include <linux/highmem.h>
  26 #include <linux/device.h>
  27 #include <linux/io.h>
  28 #include <linux/delay.h>
  29 #include <linux/netdevice.h>
  30 #include <linux/inetdevice.h>
  31 #include <linux/etherdevice.h>
  32 #include <linux/skbuff.h>
  33 #include <linux/if_vlan.h>
  34 #include <linux/in.h>
  35 #include <linux/slab.h>
  36 #include <linux/rtnetlink.h>
  37 #include <linux/netpoll.h>
  38
  39 #include <net/arp.h>
  40 #include <net/route.h>
  41 #include <net/sock.h>
  42 #include <net/pkt_sched.h>
  43 #include <net/checksum.h>
  44 #include <net/ip6_checksum.h>
  45
  46 #include "hyperv_net.h"
  47
  48 #define RING_SIZE_MIN   64
  49 #define RETRY_US_LO     5000
  50 #define RETRY_US_HI     10000
  51 #define RETRY_MAX       2000    /* >10 sec */
  52
  53 #define LINKCHANGE_INT (2 * HZ)
  54 #define VF_TAKEOVER_INT (HZ / 10)
  55
  56 static unsigned int ring_size __ro_after_init = 128;
  57 module_param(ring_size, uint, 0444);
  58 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
  59 unsigned int netvsc_ring_bytes __ro_after_init;
  60
  61 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
  62                                 NETIF_MSG_LINK | NETIF_MSG_IFUP |
  63                                 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
  64                                 NETIF_MSG_TX_ERR;
  65
  66 static int debug = -1;
  67 module_param(debug, int, 0444);
  68 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  69
  70 static LIST_HEAD(netvsc_dev_list);
  71
  72 static void netvsc_change_rx_flags(struct net_device *net, int change)
  73 {
  74         struct net_device_context *ndev_ctx = netdev_priv(net);
  75         struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
  76         int inc;
  77
  78         if (!vf_netdev)
  79                 return;
  80
  81         if (change & IFF_PROMISC) {
  82                 inc = (net->flags & IFF_PROMISC) ? 1 : -1;
  83                 dev_set_promiscuity(vf_netdev, inc);
  84         }
  85
  86         if (change & IFF_ALLMULTI) {
  87                 inc = (net->flags & IFF_ALLMULTI) ? 1 : -1;
  88                 dev_set_allmulti(vf_netdev, inc);
  89         }
  90 }
  91
  92 static void netvsc_set_rx_mode(struct net_device *net)
  93 {
  94         struct net_device_context *ndev_ctx = netdev_priv(net);
  95         struct net_device *vf_netdev;
  96         struct netvsc_device *nvdev;
  97
  98         rcu_read_lock();
  99         vf_netdev = rcu_dereference(ndev_ctx->vf_netdev);
 100         if (vf_netdev) {
 101                 dev_uc_sync(vf_netdev, net);
 102                 dev_mc_sync(vf_netdev, net);
 103         }
 104
 105         nvdev = rcu_dereference(ndev_ctx->nvdev);
 106         if (nvdev)
 107                 rndis_filter_update(nvdev);
 108         rcu_read_unlock();
 109 }
 110
 111 static int netvsc_open(struct net_device *net)
 112 {
 113         struct net_device_context *ndev_ctx = netdev_priv(net);
 114         struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
 115         struct netvsc_device *nvdev = rtnl_dereference(ndev_ctx->nvdev);
 116         struct rndis_device *rdev;
 117         int ret = 0;
 118
 119         netif_carrier_off(net);
 120
 121         /* Open up the device */
 122         ret = rndis_filter_open(nvdev);
 123         if (ret != 0) {
 124                 netdev_err(net, "unable to open device (ret %d).\n", ret);
 125                 return ret;
 126         }
 127
 128         rdev = nvdev->extension;
 129         if (!rdev->link_state) {
 130                 netif_carrier_on(net);
 131                 netif_tx_wake_all_queues(net);
 132         }
 133
 134         if (vf_netdev) {
 135                 /* Setting synthetic device up transparently sets
 136                  * slave as up. If open fails, then slave will be
 137                  * still be offline (and not used).
 138                  */
 139                 ret = dev_open(vf_netdev);
 140                 if (ret)
 141                         netdev_warn(net,
 142                                     "unable to open slave: %s: %d\n",
 143                                     vf_netdev->name, ret);
 144         }
 145         return 0;
 146 }
 147
 148 static int netvsc_wait_until_empty(struct netvsc_device *nvdev)
 149 {
 150         unsigned int retry = 0;
 151         int i;
 152
 153         /* Ensure pending bytes in ring are read */
 154         for (;;) {
 155                 u32 aread = 0;
 156
 157                 for (i = 0; i < nvdev->num_chn; i++) {
 158                         struct vmbus_channel *chn
 159                                 = nvdev->chan_table[i].channel;
 160
 161                         if (!chn)
 162                                 continue;
 163
 164                         /* make sure receive not running now */
 165                         napi_synchronize(&nvdev->chan_table[i].napi);
 166
 167                         aread = hv_get_bytes_to_read(&chn->inbound);
 168                         if (aread)
 169                                 break;
 170
 171                         aread = hv_get_bytes_to_read(&chn->outbound);
 172                         if (aread)
 173                                 break;
 174                 }
 175
 176                 if (aread == 0)
 177                         return 0;
 178
 179                 if (++retry > RETRY_MAX)
 180                         return -ETIMEDOUT;
 181
 182                 usleep_range(RETRY_US_LO, RETRY_US_HI);
 183         }
 184 }
 185
 186 static int netvsc_close(struct net_device *net)
 187 {
 188         struct net_device_context *net_device_ctx = netdev_priv(net);
 189         struct net_device *vf_netdev
 190                 = rtnl_dereference(net_device_ctx->vf_netdev);
 191         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
 192         int ret;
 193
 194         netif_tx_disable(net);
 195
 196         /* No need to close rndis filter if it is removed already */
 197         if (!nvdev)
 198                 return 0;
 199
 200         ret = rndis_filter_close(nvdev);
 201         if (ret != 0) {
 202                 netdev_err(net, "unable to close device (ret %d).\n", ret);
 203                 return ret;
 204         }
 205
 206         ret = netvsc_wait_until_empty(nvdev);
 207         if (ret)
 208                 netdev_err(net, "Ring buffer not empty after closing rndis\n");
 209
 210         if (vf_netdev)
 211                 dev_close(vf_netdev);
 212
 213         return ret;
 214 }
 215
 216 static inline void *init_ppi_data(struct rndis_message *msg,
 217                                   u32 ppi_size, u32 pkt_type)
 218 {
 219         struct rndis_packet *rndis_pkt = &msg->msg.pkt;
 220         struct rndis_per_packet_info *ppi;
 221
 222         rndis_pkt->data_offset += ppi_size;
 223         ppi = (void *)rndis_pkt + rndis_pkt->per_pkt_info_offset
 224                 + rndis_pkt->per_pkt_info_len;
 225
 226         ppi->size = ppi_size;
 227         ppi->type = pkt_type;
 228         ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
 229
 230         rndis_pkt->per_pkt_info_len += ppi_size;
 231
 232         return ppi + 1;
 233 }
 234
 235 /* Azure hosts don't support non-TCP port numbers in hashing for fragmented
 236  * packets. We can use ethtool to change UDP hash level when necessary.
 237  */
 238 static inline u32 netvsc_get_hash(
 239         struct sk_buff *skb,
 240         const struct net_device_context *ndc)
 241 {
 242         struct flow_keys flow;
 243         u32 hash, pkt_proto = 0;
 244         static u32 hashrnd __read_mostly;
 245
 246         net_get_random_once(&hashrnd, sizeof(hashrnd));
 247
 248         if (!skb_flow_dissect_flow_keys(skb, &flow, 0))
 249                 return 0;
 250
 251         switch (flow.basic.ip_proto) {
 252         case IPPROTO_TCP:
 253                 if (flow.basic.n_proto == htons(ETH_P_IP))
 254                         pkt_proto = HV_TCP4_L4HASH;
 255                 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
 256                         pkt_proto = HV_TCP6_L4HASH;
 257
 258                 break;
 259
 260         case IPPROTO_UDP:
 261                 if (flow.basic.n_proto == htons(ETH_P_IP))
 262                         pkt_proto = HV_UDP4_L4HASH;
 263                 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
 264                         pkt_proto = HV_UDP6_L4HASH;
 265
 266                 break;
 267         }
 268
 269         if (pkt_proto & ndc->l4_hash) {
 270                 return skb_get_hash(skb);
 271         } else {
 272                 if (flow.basic.n_proto == htons(ETH_P_IP))
 273                         hash = jhash2((u32 *)&flow.addrs.v4addrs, 2, hashrnd);
 274                 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
 275                         hash = jhash2((u32 *)&flow.addrs.v6addrs, 8, hashrnd);
 276                 else
 277                         hash = 0;
 278
 279                 skb_set_hash(skb, hash, PKT_HASH_TYPE_L3);
 280         }
 281
 282         return hash;
 283 }
 284
 285 static inline int netvsc_get_tx_queue(struct net_device *ndev,
 286                                       struct sk_buff *skb, int old_idx)
 287 {
 288         const struct net_device_context *ndc = netdev_priv(ndev);
 289         struct sock *sk = skb->sk;
 290         int q_idx;
 291
 292         q_idx = ndc->tx_table[netvsc_get_hash(skb, ndc) &
 293                               (VRSS_SEND_TAB_SIZE - 1)];
 294
 295         /* If queue index changed record the new value */
 296         if (q_idx != old_idx &&
 297             sk && sk_fullsock(sk) && rcu_access_pointer(sk->sk_dst_cache))
 298                 sk_tx_queue_set(sk, q_idx);
 299
 300         return q_idx;
 301 }
 302
 303 /*
 304  * Select queue for transmit.
 305  *
 306  * If a valid queue has already been assigned, then use that.
 307  * Otherwise compute tx queue based on hash and the send table.
 308  *
 309  * This is basically similar to default (__netdev_pick_tx) with the added step
 310  * of using the host send_table when no other queue has been assigned.
 311  *
 312  * TODO support XPS - but get_xps_queue not exported
 313  */
 314 static u16 netvsc_pick_tx(struct net_device *ndev, struct sk_buff *skb)
 315 {
 316         int q_idx = sk_tx_queue_get(skb->sk);
 317
 318         if (q_idx < 0 || skb->ooo_okay || q_idx >= ndev->real_num_tx_queues) {
 319                 /* If forwarding a packet, we use the recorded queue when
 320                  * available for better cache locality.
 321                  */
 322                 if (skb_rx_queue_recorded(skb))
 323                         q_idx = skb_get_rx_queue(skb);
 324                 else
 325                         q_idx = netvsc_get_tx_queue(ndev, skb, q_idx);
 326         }
 327
 328         return q_idx;
 329 }
 330
 331 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
 332                                struct net_device *sb_dev,
 333                                select_queue_fallback_t fallback)
 334 {
 335         struct net_device_context *ndc = netdev_priv(ndev);
 336         struct net_device *vf_netdev;
 337         u16 txq;
 338
 339         rcu_read_lock();
 340         vf_netdev = rcu_dereference(ndc->vf_netdev);
 341         if (vf_netdev) {
 342                 const struct net_device_ops *vf_ops = vf_netdev->netdev_ops;
 343
 344                 if (vf_ops->ndo_select_queue)
 345                         txq = vf_ops->ndo_select_queue(vf_netdev, skb,
 346                                                        sb_dev, fallback);
 347                 else
 348                         txq = fallback(vf_netdev, skb, NULL);
 349
 350                 /* Record the queue selected by VF so that it can be
 351                  * used for common case where VF has more queues than
 352                  * the synthetic device.
 353                  */
 354                 qdisc_skb_cb(skb)->slave_dev_queue_mapping = txq;
 355         } else {
 356                 txq = netvsc_pick_tx(ndev, skb);
 357         }
 358         rcu_read_unlock();
 359
 360         while (unlikely(txq >= ndev->real_num_tx_queues))
 361                 txq -= ndev->real_num_tx_queues;
 362
 363         return txq;
 364 }
 365
 366 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
 367                        struct hv_page_buffer *pb)
 368 {
 369         int j = 0;
 370
 371         /* Deal with compund pages by ignoring unused part
 372          * of the page.
 373          */
 374         page += (offset >> PAGE_SHIFT);
 375         offset &= ~PAGE_MASK;
 376
 377         while (len > 0) {
 378                 unsigned long bytes;
 379
 380                 bytes = PAGE_SIZE - offset;
 381                 if (bytes > len)
 382                         bytes = len;
 383                 pb[j].pfn = page_to_pfn(page);
 384                 pb[j].offset = offset;
 385                 pb[j].len = bytes;
 386
 387                 offset += bytes;
 388                 len -= bytes;
 389
 390                 if (offset == PAGE_SIZE && len) {
 391                         page++;
 392                         offset = 0;
 393                         j++;
 394                 }
 395         }
 396
 397         return j + 1;
 398 }
 399
 400 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
 401                            struct hv_netvsc_packet *packet,
 402                            struct hv_page_buffer *pb)
 403 {
 404         u32 slots_used = 0;
 405         char *data = skb->data;
 406         int frags = skb_shinfo(skb)->nr_frags;
 407         int i;
 408
 409         /* The packet is laid out thus:
 410          * 1. hdr: RNDIS header and PPI
 411          * 2. skb linear data
 412          * 3. skb fragment data
 413          */
 414         slots_used += fill_pg_buf(virt_to_page(hdr),
 415                                   offset_in_page(hdr),
 416                                   len, &pb[slots_used]);
 417
 418         packet->rmsg_size = len;
 419         packet->rmsg_pgcnt = slots_used;
 420
 421         slots_used += fill_pg_buf(virt_to_page(data),
 422                                 offset_in_page(data),
 423                                 skb_headlen(skb), &pb[slots_used]);
 424
 425         for (i = 0; i < frags; i++) {
 426                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 427
 428                 slots_used += fill_pg_buf(skb_frag_page(frag),
 429                                         frag->page_offset,
 430                                         skb_frag_size(frag), &pb[slots_used]);
 431         }
 432         return slots_used;
 433 }
 434
 435 static int count_skb_frag_slots(struct sk_buff *skb)
 436 {
 437         int i, frags = skb_shinfo(skb)->nr_frags;
 438         int pages = 0;
 439
 440         for (i = 0; i < frags; i++) {
 441                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 442                 unsigned long size = skb_frag_size(frag);
 443                 unsigned long offset = frag->page_offset;
 444
 445                 /* Skip unused frames from start of page */
 446                 offset &= ~PAGE_MASK;
 447                 pages += PFN_UP(offset + size);
 448         }
 449         return pages;
 450 }
 451
 452 static int netvsc_get_slots(struct sk_buff *skb)
 453 {
 454         char *data = skb->data;
 455         unsigned int offset = offset_in_page(data);
 456         unsigned int len = skb_headlen(skb);
 457         int slots;
 458         int frag_slots;
 459
 460         slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
 461         frag_slots = count_skb_frag_slots(skb);
 462         return slots + frag_slots;
 463 }
 464
 465 static u32 net_checksum_info(struct sk_buff *skb)
 466 {
 467         if (skb->protocol == htons(ETH_P_IP)) {
 468                 struct iphdr *ip = ip_hdr(skb);
 469
 470                 if (ip->protocol == IPPROTO_TCP)
 471                         return TRANSPORT_INFO_IPV4_TCP;
 472                 else if (ip->protocol == IPPROTO_UDP)
 473                         return TRANSPORT_INFO_IPV4_UDP;
 474         } else {
 475                 struct ipv6hdr *ip6 = ipv6_hdr(skb);
 476
 477                 if (ip6->nexthdr == IPPROTO_TCP)
 478                         return TRANSPORT_INFO_IPV6_TCP;
 479                 else if (ip6->nexthdr == IPPROTO_UDP)
 480                         return TRANSPORT_INFO_IPV6_UDP;
 481         }
 482
 483         return TRANSPORT_INFO_NOT_IP;
 484 }
 485
 486 /* Send skb on the slave VF device. */
 487 static int netvsc_vf_xmit(struct net_device *net, struct net_device *vf_netdev,
 488                           struct sk_buff *skb)
 489 {
 490         struct net_device_context *ndev_ctx = netdev_priv(net);
 491         unsigned int len = skb->len;
 492         int rc;
 493
 494         skb->dev = vf_netdev;
 495         skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
 496
 497         rc = dev_queue_xmit(skb);
 498         if (likely(rc == NET_XMIT_SUCCESS || rc == NET_XMIT_CN)) {
 499                 struct netvsc_vf_pcpu_stats *pcpu_stats
 500                         = this_cpu_ptr(ndev_ctx->vf_stats);
 501
 502                 u64_stats_update_begin(&pcpu_stats->syncp);
 503                 pcpu_stats->tx_packets++;
 504                 pcpu_stats->tx_bytes += len;
 505                 u64_stats_update_end(&pcpu_stats->syncp);
 506         } else {
 507                 this_cpu_inc(ndev_ctx->vf_stats->tx_dropped);
 508         }
 509
 510         return rc;
 511 }
 512
 513 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
 514 {
 515         struct net_device_context *net_device_ctx = netdev_priv(net);
 516         struct hv_netvsc_packet *packet = NULL;
 517         int ret;
 518         unsigned int num_data_pgs;
 519         struct rndis_message *rndis_msg;
 520         struct net_device *vf_netdev;
 521         u32 rndis_msg_size;
 522         u32 hash;
 523         struct hv_page_buffer pb[MAX_PAGE_BUFFER_COUNT];
 524
 525         /* if VF is present and up then redirect packets
 526          * already called with rcu_read_lock_bh
 527          */
 528         vf_netdev = rcu_dereference_bh(net_device_ctx->vf_netdev);
 529         if (vf_netdev && netif_running(vf_netdev) &&
 530             !netpoll_tx_running(net))
 531                 return netvsc_vf_xmit(net, vf_netdev, skb);
 532
 533         /* We will atmost need two pages to describe the rndis
 534          * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
 535          * of pages in a single packet. If skb is scattered around
 536          * more pages we try linearizing it.
 537          */
 538
 539         num_data_pgs = netvsc_get_slots(skb) + 2;
 540
 541         if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
 542                 ++net_device_ctx->eth_stats.tx_scattered;
 543
 544                 if (skb_linearize(skb))
 545                         goto no_memory;
 546
 547                 num_data_pgs = netvsc_get_slots(skb) + 2;
 548                 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
 549                         ++net_device_ctx->eth_stats.tx_too_big;
 550                         goto drop;
 551                 }
 552         }
 553
 554         /*
 555          * Place the rndis header in the skb head room and
 556          * the skb->cb will be used for hv_netvsc_packet
 557          * structure.
 558          */
 559         ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
 560         if (ret)
 561                 goto no_memory;
 562
 563         /* Use the skb control buffer for building up the packet */
 564         BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
 565                         FIELD_SIZEOF(struct sk_buff, cb));
 566         packet = (struct hv_netvsc_packet *)skb->cb;
 567
 568         packet->q_idx = skb_get_queue_mapping(skb);
 569
 570         packet->total_data_buflen = skb->len;
 571         packet->total_bytes = skb->len;
 572         packet->total_packets = 1;
 573
 574         rndis_msg = (struct rndis_message *)skb->head;
 575
 576         /* Add the rndis header */
 577         rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
 578         rndis_msg->msg_len = packet->total_data_buflen;
 579
 580         rndis_msg->msg.pkt = (struct rndis_packet) {
 581                 .data_offset = sizeof(struct rndis_packet),
 582                 .data_len = packet->total_data_buflen,
 583                 .per_pkt_info_offset = sizeof(struct rndis_packet),
 584         };
 585
 586         rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
 587
 588         hash = skb_get_hash_raw(skb);
 589         if (hash != 0 && net->real_num_tx_queues > 1) {
 590                 u32 *hash_info;
 591
 592                 rndis_msg_size += NDIS_HASH_PPI_SIZE;
 593                 hash_info = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
 594                                           NBL_HASH_VALUE);
 595                 *hash_info = hash;
 596         }
 597
 598         if (skb_vlan_tag_present(skb)) {
 599                 struct ndis_pkt_8021q_info *vlan;
 600
 601                 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
 602                 vlan = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
 603                                      IEEE_8021Q_INFO);
 604
 605                 vlan->value = 0;
 606                 vlan->vlanid = skb->vlan_tci & VLAN_VID_MASK;
 607                 vlan->pri = (skb->vlan_tci & VLAN_PRIO_MASK) >>
 608                                 VLAN_PRIO_SHIFT;
 609         }
 610
 611         if (skb_is_gso(skb)) {
 612                 struct ndis_tcp_lso_info *lso_info;
 613
 614                 rndis_msg_size += NDIS_LSO_PPI_SIZE;
 615                 lso_info = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
 616                                          TCP_LARGESEND_PKTINFO);
 617
 618                 lso_info->value = 0;
 619                 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
 620                 if (skb->protocol == htons(ETH_P_IP)) {
 621                         lso_info->lso_v2_transmit.ip_version =
 622                                 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
 623                         ip_hdr(skb)->tot_len = 0;
 624                         ip_hdr(skb)->check = 0;
 625                         tcp_hdr(skb)->check =
 626                                 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
 627                                                    ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 628                 } else {
 629                         lso_info->lso_v2_transmit.ip_version =
 630                                 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
 631                         ipv6_hdr(skb)->payload_len = 0;
 632                         tcp_hdr(skb)->check =
 633                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 634                                                  &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 635                 }
 636                 lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb);
 637                 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
 638         } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
 639                 if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) {
 640                         struct ndis_tcp_ip_checksum_info *csum_info;
 641
 642                         rndis_msg_size += NDIS_CSUM_PPI_SIZE;
 643                         csum_info = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
 644                                                   TCPIP_CHKSUM_PKTINFO);
 645
 646                         csum_info->value = 0;
 647                         csum_info->transmit.tcp_header_offset = skb_transport_offset(skb);
 648
 649                         if (skb->protocol == htons(ETH_P_IP)) {
 650                                 csum_info->transmit.is_ipv4 = 1;
 651
 652                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
 653                                         csum_info->transmit.tcp_checksum = 1;
 654                                 else
 655                                         csum_info->transmit.udp_checksum = 1;
 656                         } else {
 657                                 csum_info->transmit.is_ipv6 = 1;
 658
 659                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
 660                                         csum_info->transmit.tcp_checksum = 1;
 661                                 else
 662                                         csum_info->transmit.udp_checksum = 1;
 663                         }
 664                 } else {
 665                         /* Can't do offload of this type of checksum */
 666                         if (skb_checksum_help(skb))
 667                                 goto drop;
 668                 }
 669         }
 670
 671         /* Start filling in the page buffers with the rndis hdr */
 672         rndis_msg->msg_len += rndis_msg_size;
 673         packet->total_data_buflen = rndis_msg->msg_len;
 674         packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
 675                                                skb, packet, pb);
 676
 677         /* timestamp packet in software */
 678         skb_tx_timestamp(skb);
 679
 680         ret = netvsc_send(net, packet, rndis_msg, pb, skb);
 681         if (likely(ret == 0))
 682                 return NETDEV_TX_OK;
 683
 684         if (ret == -EAGAIN) {
 685                 ++net_device_ctx->eth_stats.tx_busy;
 686                 return NETDEV_TX_BUSY;
 687         }
 688
 689         if (ret == -ENOSPC)
 690                 ++net_device_ctx->eth_stats.tx_no_space;
 691
 692 drop:
 693         dev_kfree_skb_any(skb);
 694         net->stats.tx_dropped++;
 695
 696         return NETDEV_TX_OK;
 697
 698 no_memory:
 699         ++net_device_ctx->eth_stats.tx_no_memory;
 700         goto drop;
 701 }
 702
 703 /*
 704  * netvsc_linkstatus_callback - Link up/down notification
 705  */
 706 void netvsc_linkstatus_callback(struct net_device *net,
 707                                 struct rndis_message *resp)
 708 {
 709         struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
 710         struct net_device_context *ndev_ctx = netdev_priv(net);
 711         struct netvsc_reconfig *event;
 712         unsigned long flags;
 713
 714         /* Update the physical link speed when changing to another vSwitch */
 715         if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
 716                 u32 speed;
 717
 718                 speed = *(u32 *)((void *)indicate
 719                                  + indicate->status_buf_offset) / 10000;
 720                 ndev_ctx->speed = speed;
 721                 return;
 722         }
 723
 724         /* Handle these link change statuses below */
 725         if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
 726             indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
 727             indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
 728                 return;
 729
 730         if (net->reg_state != NETREG_REGISTERED)
 731                 return;
 732
 733         event = kzalloc(sizeof(*event), GFP_ATOMIC);
 734         if (!event)
 735                 return;
 736         event->event = indicate->status;
 737
 738         spin_lock_irqsave(&ndev_ctx->lock, flags);
 739         list_add_tail(&event->list, &ndev_ctx->reconfig_events);
 740         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
 741
 742         schedule_delayed_work(&ndev_ctx->dwork, 0);
 743 }
 744
 745 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
 746                                              struct napi_struct *napi,
 747                                              const struct ndis_tcp_ip_checksum_info *csum_info,
 748                                              const struct ndis_pkt_8021q_info *vlan,
 749                                              void *data, u32 buflen)
 750 {
 751         struct sk_buff *skb;
 752
 753         skb = napi_alloc_skb(napi, buflen);
 754         if (!skb)
 755                 return skb;
 756
 757         /*
 758          * Copy to skb. This copy is needed here since the memory pointed by
 759          * hv_netvsc_packet cannot be deallocated
 760          */
 761         skb_put_data(skb, data, buflen);
 762
 763         skb->protocol = eth_type_trans(skb, net);
 764
 765         /* skb is already created with CHECKSUM_NONE */
 766         skb_checksum_none_assert(skb);
 767
 768         /*
 769          * In Linux, the IP checksum is always checked.
 770          * Do L4 checksum offload if enabled and present.
 771          */
 772         if (csum_info && (net->features & NETIF_F_RXCSUM)) {
 773                 if (csum_info->receive.tcp_checksum_succeeded ||
 774                     csum_info->receive.udp_checksum_succeeded)
 775                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 776         }
 777
 778         if (vlan) {
 779                 u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT);
 780
 781                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 782                                        vlan_tci);
 783         }
 784
 785         return skb;
 786 }
 787
 788 /*
 789  * netvsc_recv_callback -  Callback when we receive a packet from the
 790  * "wire" on the specified device.
 791  */
 792 int netvsc_recv_callback(struct net_device *net,
 793                          struct netvsc_device *net_device,
 794                          struct vmbus_channel *channel,
 795                          void  *data, u32 len,
 796                          const struct ndis_tcp_ip_checksum_info *csum_info,
 797                          const struct ndis_pkt_8021q_info *vlan)
 798 {
 799         struct net_device_context *net_device_ctx = netdev_priv(net);
 800         u16 q_idx = channel->offermsg.offer.sub_channel_index;
 801         struct netvsc_channel *nvchan = &net_device->chan_table[q_idx];
 802         struct sk_buff *skb;
 803         struct netvsc_stats *rx_stats;
 804
 805         if (net->reg_state != NETREG_REGISTERED)
 806                 return NVSP_STAT_FAIL;
 807
 808         /* Allocate a skb - TODO direct I/O to pages? */
 809         skb = netvsc_alloc_recv_skb(net, &nvchan->napi,
 810                                     csum_info, vlan, data, len);
 811         if (unlikely(!skb)) {
 812                 ++net_device_ctx->eth_stats.rx_no_memory;
 813                 rcu_read_unlock();
 814                 return NVSP_STAT_FAIL;
 815         }
 816
 817         skb_record_rx_queue(skb, q_idx);
 818
 819         /*
 820          * Even if injecting the packet, record the statistics
 821          * on the synthetic device because modifying the VF device
 822          * statistics will not work correctly.
 823          */
 824         rx_stats = &nvchan->rx_stats;
 825         u64_stats_update_begin(&rx_stats->syncp);
 826         rx_stats->packets++;
 827         rx_stats->bytes += len;
 828
 829         if (skb->pkt_type == PACKET_BROADCAST)
 830                 ++rx_stats->broadcast;
 831         else if (skb->pkt_type == PACKET_MULTICAST)
 832                 ++rx_stats->multicast;
 833         u64_stats_update_end(&rx_stats->syncp);
 834
 835         napi_gro_receive(&nvchan->napi, skb);
 836         return NVSP_STAT_SUCCESS;
 837 }
 838
 839 static void netvsc_get_drvinfo(struct net_device *net,
 840                                struct ethtool_drvinfo *info)
 841 {
 842         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
 843         strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
 844 }
 845
 846 static void netvsc_get_channels(struct net_device *net,
 847                                 struct ethtool_channels *channel)
 848 {
 849         struct net_device_context *net_device_ctx = netdev_priv(net);
 850         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
 851
 852         if (nvdev) {
 853                 channel->max_combined   = nvdev->max_chn;
 854                 channel->combined_count = nvdev->num_chn;
 855         }
 856 }
 857
 858 static int netvsc_detach(struct net_device *ndev,
 859                          struct netvsc_device *nvdev)
 860 {
 861         struct net_device_context *ndev_ctx = netdev_priv(ndev);
 862         struct hv_device *hdev = ndev_ctx->device_ctx;
 863         int ret;
 864
 865         /* Don't try continuing to try and setup sub channels */
 866         if (cancel_work_sync(&nvdev->subchan_work))
 867                 nvdev->num_chn = 1;
 868
 869         /* If device was up (receiving) then shutdown */
 870         if (netif_running(ndev)) {
 871                 netif_tx_disable(ndev);
 872
 873                 ret = rndis_filter_close(nvdev);
 874                 if (ret) {
 875                         netdev_err(ndev,
 876                                    "unable to close device (ret %d).\n", ret);
 877                         return ret;
 878                 }
 879
 880                 ret = netvsc_wait_until_empty(nvdev);
 881                 if (ret) {
 882                         netdev_err(ndev,
 883                                    "Ring buffer not empty after closing rndis\n");
 884                         return ret;
 885                 }
 886         }
 887
 888         netif_device_detach(ndev);
 889
 890         rndis_filter_device_remove(hdev, nvdev);
 891
 892         return 0;
 893 }
 894
 895 static int netvsc_attach(struct net_device *ndev,
 896                          struct netvsc_device_info *dev_info)
 897 {
 898         struct net_device_context *ndev_ctx = netdev_priv(ndev);
 899         struct hv_device *hdev = ndev_ctx->device_ctx;
 900         struct netvsc_device *nvdev;
 901         struct rndis_device *rdev;
 902         int ret;
 903
 904         nvdev = rndis_filter_device_add(hdev, dev_info);
 905         if (IS_ERR(nvdev))
 906                 return PTR_ERR(nvdev);
 907
 908         if (nvdev->num_chn > 1) {
 909                 ret = rndis_set_subchannel(ndev, nvdev);
 910
 911                 /* if unavailable, just proceed with one queue */
 912                 if (ret) {
 913                         nvdev->max_chn = 1;
 914                         nvdev->num_chn = 1;
 915                 }
 916         }
 917
 918         /* In any case device is now ready */
 919         netif_device_attach(ndev);
 920
 921         /* Note: enable and attach happen when sub-channels setup */
 922         netif_carrier_off(ndev);
 923
 924         if (netif_running(ndev)) {
 925                 ret = rndis_filter_open(nvdev);
 926                 if (ret)
 927                         return ret;
 928
 929                 rdev = nvdev->extension;
 930                 if (!rdev->link_state)
 931                         netif_carrier_on(ndev);
 932         }
 933
 934         return 0;
 935 }
 936
 937 static int netvsc_set_channels(struct net_device *net,
 938                                struct ethtool_channels *channels)
 939 {
 940         struct net_device_context *net_device_ctx = netdev_priv(net);
 941         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
 942         unsigned int orig, count = channels->combined_count;
 943         struct netvsc_device_info device_info;
 944         int ret;
 945
 946         /* We do not support separate count for rx, tx, or other */
 947         if (count == 0 ||
 948             channels->rx_count || channels->tx_count || channels->other_count)
 949                 return -EINVAL;
 950
 951         if (!nvdev || nvdev->destroy)
 952                 return -ENODEV;
 953
 954         if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
 955                 return -EINVAL;
 956
 957         if (count > nvdev->max_chn)
 958                 return -EINVAL;
 959
 960         orig = nvdev->num_chn;
 961
 962         memset(&device_info, 0, sizeof(device_info));
 963         device_info.num_chn = count;
 964         device_info.send_sections = nvdev->send_section_cnt;
 965         device_info.send_section_size = nvdev->send_section_size;
 966         device_info.recv_sections = nvdev->recv_section_cnt;
 967         device_info.recv_section_size = nvdev->recv_section_size;
 968
 969         ret = netvsc_detach(net, nvdev);
 970         if (ret)
 971                 return ret;
 972
 973         ret = netvsc_attach(net, &device_info);
 974         if (ret) {
 975                 device_info.num_chn = orig;
 976                 if (netvsc_attach(net, &device_info))
 977                         netdev_err(net, "restoring channel setting failed\n");
 978         }
 979
 980         return ret;
 981 }
 982
 983 static bool
 984 netvsc_validate_ethtool_ss_cmd(const struct ethtool_link_ksettings *cmd)
 985 {
 986         struct ethtool_link_ksettings diff1 = *cmd;
 987         struct ethtool_link_ksettings diff2 = {};
 988
 989         diff1.base.speed = 0;
 990         diff1.base.duplex = 0;
 991         /* advertising and cmd are usually set */
 992         ethtool_link_ksettings_zero_link_mode(&diff1, advertising);
 993         diff1.base.cmd = 0;
 994         /* We set port to PORT_OTHER */
 995         diff2.base.port = PORT_OTHER;
 996
 997         return !memcmp(&diff1, &diff2, sizeof(diff1));
 998 }
 999
1000 static void netvsc_init_settings(struct net_device *dev)
1001 {
1002         struct net_device_context *ndc = netdev_priv(dev);
1003
1004         ndc->l4_hash = HV_DEFAULT_L4HASH;
1005
1006         ndc->speed = SPEED_UNKNOWN;
1007         ndc->duplex = DUPLEX_FULL;
1008 }
1009
1010 static int netvsc_get_link_ksettings(struct net_device *dev,
1011                                      struct ethtool_link_ksettings *cmd)
1012 {
1013         struct net_device_context *ndc = netdev_priv(dev);
1014
1015         cmd->base.speed = ndc->speed;
1016         cmd->base.duplex = ndc->duplex;
1017         cmd->base.port = PORT_OTHER;
1018
1019         return 0;
1020 }
1021
1022 static int netvsc_set_link_ksettings(struct net_device *dev,
1023                                      const struct ethtool_link_ksettings *cmd)
1024 {
1025         struct net_device_context *ndc = netdev_priv(dev);
1026         u32 speed;
1027
1028         speed = cmd->base.speed;
1029         if (!ethtool_validate_speed(speed) ||
1030             !ethtool_validate_duplex(cmd->base.duplex) ||
1031             !netvsc_validate_ethtool_ss_cmd(cmd))
1032                 return -EINVAL;
1033
1034         ndc->speed = speed;
1035         ndc->duplex = cmd->base.duplex;
1036
1037         return 0;
1038 }
1039
1040 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
1041 {
1042         struct net_device_context *ndevctx = netdev_priv(ndev);
1043         struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1044         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1045         int orig_mtu = ndev->mtu;
1046         struct netvsc_device_info device_info;
1047         int ret = 0;
1048
1049         if (!nvdev || nvdev->destroy)
1050                 return -ENODEV;
1051
1052         /* Change MTU of underlying VF netdev first. */
1053         if (vf_netdev) {
1054                 ret = dev_set_mtu(vf_netdev, mtu);
1055                 if (ret)
1056                         return ret;
1057         }
1058
1059         memset(&device_info, 0, sizeof(device_info));
1060         device_info.num_chn = nvdev->num_chn;
1061         device_info.send_sections = nvdev->send_section_cnt;
1062         device_info.send_section_size = nvdev->send_section_size;
1063         device_info.recv_sections = nvdev->recv_section_cnt;
1064         device_info.recv_section_size = nvdev->recv_section_size;
1065
1066         ret = netvsc_detach(ndev, nvdev);
1067         if (ret)
1068                 goto rollback_vf;
1069
1070         ndev->mtu = mtu;
1071
1072         ret = netvsc_attach(ndev, &device_info);
1073         if (ret)
1074                 goto rollback;
1075
1076         return 0;
1077
1078 rollback:
1079         /* Attempt rollback to original MTU */
1080         ndev->mtu = orig_mtu;
1081
1082         if (netvsc_attach(ndev, &device_info))
1083                 netdev_err(ndev, "restoring mtu failed\n");
1084 rollback_vf:
1085         if (vf_netdev)
1086                 dev_set_mtu(vf_netdev, orig_mtu);
1087
1088         return ret;
1089 }
1090
1091 static void netvsc_get_vf_stats(struct net_device *net,
1092                                 struct netvsc_vf_pcpu_stats *tot)
1093 {
1094         struct net_device_context *ndev_ctx = netdev_priv(net);
1095         int i;
1096
1097         memset(tot, 0, sizeof(*tot));
1098
1099         for_each_possible_cpu(i) {
1100                 const struct netvsc_vf_pcpu_stats *stats
1101                         = per_cpu_ptr(ndev_ctx->vf_stats, i);
1102                 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1103                 unsigned int start;
1104
1105                 do {
1106                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1107                         rx_packets = stats->rx_packets;
1108                         tx_packets = stats->tx_packets;
1109                         rx_bytes = stats->rx_bytes;
1110                         tx_bytes = stats->tx_bytes;
1111                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1112
1113                 tot->rx_packets += rx_packets;
1114                 tot->tx_packets += tx_packets;
1115                 tot->rx_bytes   += rx_bytes;
1116                 tot->tx_bytes   += tx_bytes;
1117                 tot->tx_dropped += stats->tx_dropped;
1118         }
1119 }
1120
1121 static void netvsc_get_pcpu_stats(struct net_device *net,
1122                                   struct netvsc_ethtool_pcpu_stats *pcpu_tot)
1123 {
1124         struct net_device_context *ndev_ctx = netdev_priv(net);
1125         struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
1126         int i;
1127
1128         /* fetch percpu stats of vf */
1129         for_each_possible_cpu(i) {
1130                 const struct netvsc_vf_pcpu_stats *stats =
1131                         per_cpu_ptr(ndev_ctx->vf_stats, i);
1132                 struct netvsc_ethtool_pcpu_stats *this_tot = &pcpu_tot[i];
1133                 unsigned int start;
1134
1135                 do {
1136                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1137                         this_tot->vf_rx_packets = stats->rx_packets;
1138                         this_tot->vf_tx_packets = stats->tx_packets;
1139                         this_tot->vf_rx_bytes = stats->rx_bytes;
1140                         this_tot->vf_tx_bytes = stats->tx_bytes;
1141                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1142                 this_tot->rx_packets = this_tot->vf_rx_packets;
1143                 this_tot->tx_packets = this_tot->vf_tx_packets;
1144                 this_tot->rx_bytes   = this_tot->vf_rx_bytes;
1145                 this_tot->tx_bytes   = this_tot->vf_tx_bytes;
1146         }
1147
1148         /* fetch percpu stats of netvsc */
1149         for (i = 0; i < nvdev->num_chn; i++) {
1150                 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1151                 const struct netvsc_stats *stats;
1152                 struct netvsc_ethtool_pcpu_stats *this_tot =
1153                         &pcpu_tot[nvchan->channel->target_cpu];
1154                 u64 packets, bytes;
1155                 unsigned int start;
1156
1157                 stats = &nvchan->tx_stats;
1158                 do {
1159                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1160                         packets = stats->packets;
1161                         bytes = stats->bytes;
1162                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1163
1164                 this_tot->tx_bytes      += bytes;
1165                 this_tot->tx_packets    += packets;
1166
1167                 stats = &nvchan->rx_stats;
1168                 do {
1169                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1170                         packets = stats->packets;
1171                         bytes = stats->bytes;
1172                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1173
1174                 this_tot->rx_bytes      += bytes;
1175                 this_tot->rx_packets    += packets;
1176         }
1177 }
1178
1179 static void netvsc_get_stats64(struct net_device *net,
1180                                struct rtnl_link_stats64 *t)
1181 {
1182         struct net_device_context *ndev_ctx = netdev_priv(net);
1183         struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
1184         struct netvsc_vf_pcpu_stats vf_tot;
1185         int i;
1186
1187         if (!nvdev)
1188                 return;
1189
1190         netdev_stats_to_stats64(t, &net->stats);
1191
1192         netvsc_get_vf_stats(net, &vf_tot);
1193         t->rx_packets += vf_tot.rx_packets;
1194         t->tx_packets += vf_tot.tx_packets;
1195         t->rx_bytes   += vf_tot.rx_bytes;
1196         t->tx_bytes   += vf_tot.tx_bytes;
1197         t->tx_dropped += vf_tot.tx_dropped;
1198
1199         for (i = 0; i < nvdev->num_chn; i++) {
1200                 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1201                 const struct netvsc_stats *stats;
1202                 u64 packets, bytes, multicast;
1203                 unsigned int start;
1204
1205                 stats = &nvchan->tx_stats;
1206                 do {
1207                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1208                         packets = stats->packets;
1209                         bytes = stats->bytes;
1210                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1211
1212                 t->tx_bytes     += bytes;
1213                 t->tx_packets   += packets;
1214
1215                 stats = &nvchan->rx_stats;
1216                 do {
1217                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1218                         packets = stats->packets;
1219                         bytes = stats->bytes;
1220                         multicast = stats->multicast + stats->broadcast;
1221                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1222
1223                 t->rx_bytes     += bytes;
1224                 t->rx_packets   += packets;
1225                 t->multicast    += multicast;
1226         }
1227 }
1228
1229 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
1230 {
1231         struct net_device_context *ndc = netdev_priv(ndev);
1232         struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1233         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1234         struct sockaddr *addr = p;
1235         int err;
1236
1237         err = eth_prepare_mac_addr_change(ndev, p);
1238         if (err)
1239                 return err;
1240
1241         if (!nvdev)
1242                 return -ENODEV;
1243
1244         if (vf_netdev) {
1245                 err = dev_set_mac_address(vf_netdev, addr);
1246                 if (err)
1247                         return err;
1248         }
1249
1250         err = rndis_filter_set_device_mac(nvdev, addr->sa_data);
1251         if (!err) {
1252                 eth_commit_mac_addr_change(ndev, p);
1253         } else if (vf_netdev) {
1254                 /* rollback change on VF */
1255                 memcpy(addr->sa_data, ndev->dev_addr, ETH_ALEN);
1256                 dev_set_mac_address(vf_netdev, addr);
1257         }
1258
1259         return err;
1260 }
1261
1262 static const struct {
1263         char name[ETH_GSTRING_LEN];
1264         u16 offset;
1265 } netvsc_stats[] = {
1266         { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
1267         { "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
1268         { "tx_no_space",  offsetof(struct netvsc_ethtool_stats, tx_no_space) },
1269         { "tx_too_big",   offsetof(struct netvsc_ethtool_stats, tx_too_big) },
1270         { "tx_busy",      offsetof(struct netvsc_ethtool_stats, tx_busy) },
1271         { "tx_send_full", offsetof(struct netvsc_ethtool_stats, tx_send_full) },
1272         { "rx_comp_busy", offsetof(struct netvsc_ethtool_stats, rx_comp_busy) },
1273         { "rx_no_memory", offsetof(struct netvsc_ethtool_stats, rx_no_memory) },
1274         { "stop_queue", offsetof(struct netvsc_ethtool_stats, stop_queue) },
1275         { "wake_queue", offsetof(struct netvsc_ethtool_stats, wake_queue) },
1276 }, pcpu_stats[] = {
1277         { "cpu%u_rx_packets",
1278                 offsetof(struct netvsc_ethtool_pcpu_stats, rx_packets) },
1279         { "cpu%u_rx_bytes",
1280                 offsetof(struct netvsc_ethtool_pcpu_stats, rx_bytes) },
1281         { "cpu%u_tx_packets",
1282                 offsetof(struct netvsc_ethtool_pcpu_stats, tx_packets) },
1283         { "cpu%u_tx_bytes",
1284                 offsetof(struct netvsc_ethtool_pcpu_stats, tx_bytes) },
1285         { "cpu%u_vf_rx_packets",
1286                 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_packets) },
1287         { "cpu%u_vf_rx_bytes",
1288                 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_bytes) },
1289         { "cpu%u_vf_tx_packets",
1290                 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_packets) },
1291         { "cpu%u_vf_tx_bytes",
1292                 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_bytes) },
1293 }, vf_stats[] = {
1294         { "vf_rx_packets", offsetof(struct netvsc_vf_pcpu_stats, rx_packets) },
1295         { "vf_rx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, rx_bytes) },
1296         { "vf_tx_packets", offsetof(struct netvsc_vf_pcpu_stats, tx_packets) },
1297         { "vf_tx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, tx_bytes) },
1298         { "vf_tx_dropped", offsetof(struct netvsc_vf_pcpu_stats, tx_dropped) },
1299 };
1300
1301 #define NETVSC_GLOBAL_STATS_LEN ARRAY_SIZE(netvsc_stats)
1302 #define NETVSC_VF_STATS_LEN     ARRAY_SIZE(vf_stats)
1303
1304 /* statistics per queue (rx/tx packets/bytes) */
1305 #define NETVSC_PCPU_STATS_LEN (num_present_cpus() * ARRAY_SIZE(pcpu_stats))
1306
1307 /* 4 statistics per queue (rx/tx packets/bytes) */
1308 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 4)
1309
1310 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
1311 {
1312         struct net_device_context *ndc = netdev_priv(dev);
1313         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1314
1315         if (!nvdev)
1316                 return -ENODEV;
1317
1318         switch (string_set) {
1319         case ETH_SS_STATS:
1320                 return NETVSC_GLOBAL_STATS_LEN
1321                         + NETVSC_VF_STATS_LEN
1322                         + NETVSC_QUEUE_STATS_LEN(nvdev)
1323                         + NETVSC_PCPU_STATS_LEN;
1324         default:
1325                 return -EINVAL;
1326         }
1327 }
1328
1329 static void netvsc_get_ethtool_stats(struct net_device *dev,
1330                                      struct ethtool_stats *stats, u64 *data)
1331 {
1332         struct net_device_context *ndc = netdev_priv(dev);
1333         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1334         const void *nds = &ndc->eth_stats;
1335         const struct netvsc_stats *qstats;
1336         struct netvsc_vf_pcpu_stats sum;
1337         struct netvsc_ethtool_pcpu_stats *pcpu_sum;
1338         unsigned int start;
1339         u64 packets, bytes;
1340         int i, j, cpu;
1341
1342         if (!nvdev)
1343                 return;
1344
1345         for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
1346                 data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1347
1348         netvsc_get_vf_stats(dev, &sum);
1349         for (j = 0; j < NETVSC_VF_STATS_LEN; j++)
1350                 data[i++] = *(u64 *)((void *)&sum + vf_stats[j].offset);
1351
1352         for (j = 0; j < nvdev->num_chn; j++) {
1353                 qstats = &nvdev->chan_table[j].tx_stats;
1354
1355                 do {
1356                         start = u64_stats_fetch_begin_irq(&qstats->syncp);
1357                         packets = qstats->packets;
1358                         bytes = qstats->bytes;
1359                 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1360                 data[i++] = packets;
1361                 data[i++] = bytes;
1362
1363                 qstats = &nvdev->chan_table[j].rx_stats;
1364                 do {
1365                         start = u64_stats_fetch_begin_irq(&qstats->syncp);
1366                         packets = qstats->packets;
1367                         bytes = qstats->bytes;
1368                 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1369                 data[i++] = packets;
1370                 data[i++] = bytes;
1371         }
1372
1373         pcpu_sum = kvmalloc_array(num_possible_cpus(),
1374                                   sizeof(struct netvsc_ethtool_pcpu_stats),
1375                                   GFP_KERNEL);
1376         netvsc_get_pcpu_stats(dev, pcpu_sum);
1377         for_each_present_cpu(cpu) {
1378                 struct netvsc_ethtool_pcpu_stats *this_sum = &pcpu_sum[cpu];
1379
1380                 for (j = 0; j < ARRAY_SIZE(pcpu_stats); j++)
1381                         data[i++] = *(u64 *)((void *)this_sum
1382                                              + pcpu_stats[j].offset);
1383         }
1384         kvfree(pcpu_sum);
1385 }
1386
1387 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1388 {
1389         struct net_device_context *ndc = netdev_priv(dev);
1390         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1391         u8 *p = data;
1392         int i, cpu;
1393
1394         if (!nvdev)
1395                 return;
1396
1397         switch (stringset) {
1398         case ETH_SS_STATS:
1399                 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++) {
1400                         memcpy(p, netvsc_stats[i].name, ETH_GSTRING_LEN);
1401                         p += ETH_GSTRING_LEN;
1402                 }
1403
1404                 for (i = 0; i < ARRAY_SIZE(vf_stats); i++) {
1405                         memcpy(p, vf_stats[i].name, ETH_GSTRING_LEN);
1406                         p += ETH_GSTRING_LEN;
1407                 }
1408
1409                 for (i = 0; i < nvdev->num_chn; i++) {
1410                         sprintf(p, "tx_queue_%u_packets", i);
1411                         p += ETH_GSTRING_LEN;
1412                         sprintf(p, "tx_queue_%u_bytes", i);
1413                         p += ETH_GSTRING_LEN;
1414                         sprintf(p, "rx_queue_%u_packets", i);
1415                         p += ETH_GSTRING_LEN;
1416                         sprintf(p, "rx_queue_%u_bytes", i);
1417                         p += ETH_GSTRING_LEN;
1418                 }
1419
1420                 for_each_present_cpu(cpu) {
1421                         for (i = 0; i < ARRAY_SIZE(pcpu_stats); i++) {
1422                                 sprintf(p, pcpu_stats[i].name, cpu);
1423                                 p += ETH_GSTRING_LEN;
1424                         }
1425                 }
1426
1427                 break;
1428         }
1429 }
1430
1431 static int
1432 netvsc_get_rss_hash_opts(struct net_device_context *ndc,
1433                          struct ethtool_rxnfc *info)
1434 {
1435         const u32 l4_flag = RXH_L4_B_0_1 | RXH_L4_B_2_3;
1436
1437         info->data = RXH_IP_SRC | RXH_IP_DST;
1438
1439         switch (info->flow_type) {
1440         case TCP_V4_FLOW:
1441                 if (ndc->l4_hash & HV_TCP4_L4HASH)
1442                         info->data |= l4_flag;
1443
1444                 break;
1445
1446         case TCP_V6_FLOW:
1447                 if (ndc->l4_hash & HV_TCP6_L4HASH)
1448                         info->data |= l4_flag;
1449
1450                 break;
1451
1452         case UDP_V4_FLOW:
1453                 if (ndc->l4_hash & HV_UDP4_L4HASH)
1454                         info->data |= l4_flag;
1455
1456                 break;
1457
1458         case UDP_V6_FLOW:
1459                 if (ndc->l4_hash & HV_UDP6_L4HASH)
1460                         info->data |= l4_flag;
1461
1462                 break;
1463
1464         case IPV4_FLOW:
1465         case IPV6_FLOW:
1466                 break;
1467         default:
1468                 info->data = 0;
1469                 break;
1470         }
1471
1472         return 0;
1473 }
1474
1475 static int
1476 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1477                  u32 *rules)
1478 {
1479         struct net_device_context *ndc = netdev_priv(dev);
1480         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1481
1482         if (!nvdev)
1483                 return -ENODEV;
1484
1485         switch (info->cmd) {
1486         case ETHTOOL_GRXRINGS:
1487                 info->data = nvdev->num_chn;
1488                 return 0;
1489
1490         case ETHTOOL_GRXFH:
1491                 return netvsc_get_rss_hash_opts(ndc, info);
1492         }
1493         return -EOPNOTSUPP;
1494 }
1495
1496 static int netvsc_set_rss_hash_opts(struct net_device_context *ndc,
1497                                     struct ethtool_rxnfc *info)
1498 {
1499         if (info->data == (RXH_IP_SRC | RXH_IP_DST |
1500                            RXH_L4_B_0_1 | RXH_L4_B_2_3)) {
1501                 switch (info->flow_type) {
1502                 case TCP_V4_FLOW:
1503                         ndc->l4_hash |= HV_TCP4_L4HASH;
1504                         break;
1505
1506                 case TCP_V6_FLOW:
1507                         ndc->l4_hash |= HV_TCP6_L4HASH;
1508                         break;
1509
1510                 case UDP_V4_FLOW:
1511                         ndc->l4_hash |= HV_UDP4_L4HASH;
1512                         break;
1513
1514                 case UDP_V6_FLOW:
1515                         ndc->l4_hash |= HV_UDP6_L4HASH;
1516                         break;
1517
1518                 default:
1519                         return -EOPNOTSUPP;
1520                 }
1521
1522                 return 0;
1523         }
1524
1525         if (info->data == (RXH_IP_SRC | RXH_IP_DST)) {
1526                 switch (info->flow_type) {
1527                 case TCP_V4_FLOW:
1528                         ndc->l4_hash &= ~HV_TCP4_L4HASH;
1529                         break;
1530
1531                 case TCP_V6_FLOW:
1532                         ndc->l4_hash &= ~HV_TCP6_L4HASH;
1533                         break;
1534
1535                 case UDP_V4_FLOW:
1536                         ndc->l4_hash &= ~HV_UDP4_L4HASH;
1537                         break;
1538
1539                 case UDP_V6_FLOW:
1540                         ndc->l4_hash &= ~HV_UDP6_L4HASH;
1541                         break;
1542
1543                 default:
1544                         return -EOPNOTSUPP;
1545                 }
1546
1547                 return 0;
1548         }
1549
1550         return -EOPNOTSUPP;
1551 }
1552
1553 static int
1554 netvsc_set_rxnfc(struct net_device *ndev, struct ethtool_rxnfc *info)
1555 {
1556         struct net_device_context *ndc = netdev_priv(ndev);
1557
1558         if (info->cmd == ETHTOOL_SRXFH)
1559                 return netvsc_set_rss_hash_opts(ndc, info);
1560
1561         return -EOPNOTSUPP;
1562 }
1563
1564 #ifdef CONFIG_NET_POLL_CONTROLLER
1565 static void netvsc_poll_controller(struct net_device *dev)
1566 {
1567         struct net_device_context *ndc = netdev_priv(dev);
1568         struct netvsc_device *ndev;
1569         int i;
1570
1571         rcu_read_lock();
1572         ndev = rcu_dereference(ndc->nvdev);
1573         if (ndev) {
1574                 for (i = 0; i < ndev->num_chn; i++) {
1575                         struct netvsc_channel *nvchan = &ndev->chan_table[i];
1576
1577                         napi_schedule(&nvchan->napi);
1578                 }
1579         }
1580         rcu_read_unlock();
1581 }
1582 #endif
1583
1584 static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1585 {
1586         return NETVSC_HASH_KEYLEN;
1587 }
1588
1589 static u32 netvsc_rss_indir_size(struct net_device *dev)
1590 {
1591         return ITAB_NUM;
1592 }
1593
1594 static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key,
1595                            u8 *hfunc)
1596 {
1597         struct net_device_context *ndc = netdev_priv(dev);
1598         struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1599         struct rndis_device *rndis_dev;
1600         int i;
1601
1602         if (!ndev)
1603                 return -ENODEV;
1604
1605         if (hfunc)
1606                 *hfunc = ETH_RSS_HASH_TOP;      /* Toeplitz */
1607
1608         rndis_dev = ndev->extension;
1609         if (indir) {
1610                 for (i = 0; i < ITAB_NUM; i++)
1611                         indir[i] = rndis_dev->rx_table[i];
1612         }
1613
1614         if (key)
1615                 memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1616
1617         return 0;
1618 }
1619
1620 static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
1621                            const u8 *key, const u8 hfunc)
1622 {
1623         struct net_device_context *ndc = netdev_priv(dev);
1624         struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1625         struct rndis_device *rndis_dev;
1626         int i;
1627
1628         if (!ndev)
1629                 return -ENODEV;
1630
1631         if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
1632                 return -EOPNOTSUPP;
1633
1634         rndis_dev = ndev->extension;
1635         if (indir) {
1636                 for (i = 0; i < ITAB_NUM; i++)
1637                         if (indir[i] >= ndev->num_chn)
1638                                 return -EINVAL;
1639
1640                 for (i = 0; i < ITAB_NUM; i++)
1641                         rndis_dev->rx_table[i] = indir[i];
1642         }
1643
1644         if (!key) {
1645                 if (!indir)
1646                         return 0;
1647
1648                 key = rndis_dev->rss_key;
1649         }
1650
1651         return rndis_filter_set_rss_param(rndis_dev, key);
1652 }
1653
1654 /* Hyper-V RNDIS protocol does not have ring in the HW sense.
1655  * It does have pre-allocated receive area which is divided into sections.
1656  */
1657 static void __netvsc_get_ringparam(struct netvsc_device *nvdev,
1658                                    struct ethtool_ringparam *ring)
1659 {
1660         u32 max_buf_size;
1661
1662         ring->rx_pending = nvdev->recv_section_cnt;
1663         ring->tx_pending = nvdev->send_section_cnt;
1664
1665         if (nvdev->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
1666                 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
1667         else
1668                 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
1669
1670         ring->rx_max_pending = max_buf_size / nvdev->recv_section_size;
1671         ring->tx_max_pending = NETVSC_SEND_BUFFER_SIZE
1672                 / nvdev->send_section_size;
1673 }
1674
1675 static void netvsc_get_ringparam(struct net_device *ndev,
1676                                  struct ethtool_ringparam *ring)
1677 {
1678         struct net_device_context *ndevctx = netdev_priv(ndev);
1679         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1680
1681         if (!nvdev)
1682                 return;
1683
1684         __netvsc_get_ringparam(nvdev, ring);
1685 }
1686
1687 static int netvsc_set_ringparam(struct net_device *ndev,
1688                                 struct ethtool_ringparam *ring)
1689 {
1690         struct net_device_context *ndevctx = netdev_priv(ndev);
1691         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1692         struct netvsc_device_info device_info;
1693         struct ethtool_ringparam orig;
1694         u32 new_tx, new_rx;
1695         int ret = 0;
1696
1697         if (!nvdev || nvdev->destroy)
1698                 return -ENODEV;
1699
1700         memset(&orig, 0, sizeof(orig));
1701         __netvsc_get_ringparam(nvdev, &orig);
1702
1703         new_tx = clamp_t(u32, ring->tx_pending,
1704                          NETVSC_MIN_TX_SECTIONS, orig.tx_max_pending);
1705         new_rx = clamp_t(u32, ring->rx_pending,
1706                          NETVSC_MIN_RX_SECTIONS, orig.rx_max_pending);
1707
1708         if (new_tx == orig.tx_pending &&
1709             new_rx == orig.rx_pending)
1710                 return 0;        /* no change */
1711
1712         memset(&device_info, 0, sizeof(device_info));
1713         device_info.num_chn = nvdev->num_chn;
1714         device_info.send_sections = new_tx;
1715         device_info.send_section_size = nvdev->send_section_size;
1716         device_info.recv_sections = new_rx;
1717         device_info.recv_section_size = nvdev->recv_section_size;
1718
1719         ret = netvsc_detach(ndev, nvdev);
1720         if (ret)
1721                 return ret;
1722
1723         ret = netvsc_attach(ndev, &device_info);
1724         if (ret) {
1725                 device_info.send_sections = orig.tx_pending;
1726                 device_info.recv_sections = orig.rx_pending;
1727
1728                 if (netvsc_attach(ndev, &device_info))
1729                         netdev_err(ndev, "restoring ringparam failed");
1730         }
1731
1732         return ret;
1733 }
1734
1735 static u32 netvsc_get_msglevel(struct net_device *ndev)
1736 {
1737         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1738
1739         return ndev_ctx->msg_enable;
1740 }
1741
1742 static void netvsc_set_msglevel(struct net_device *ndev, u32 val)
1743 {
1744         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1745
1746         ndev_ctx->msg_enable = val;
1747 }
1748
1749 static const struct ethtool_ops ethtool_ops = {
1750         .get_drvinfo    = netvsc_get_drvinfo,
1751         .get_msglevel   = netvsc_get_msglevel,
1752         .set_msglevel   = netvsc_set_msglevel,
1753         .get_link       = ethtool_op_get_link,
1754         .get_ethtool_stats = netvsc_get_ethtool_stats,
1755         .get_sset_count = netvsc_get_sset_count,
1756         .get_strings    = netvsc_get_strings,
1757         .get_channels   = netvsc_get_channels,
1758         .set_channels   = netvsc_set_channels,
1759         .get_ts_info    = ethtool_op_get_ts_info,
1760         .get_rxnfc      = netvsc_get_rxnfc,
1761         .set_rxnfc      = netvsc_set_rxnfc,
1762         .get_rxfh_key_size = netvsc_get_rxfh_key_size,
1763         .get_rxfh_indir_size = netvsc_rss_indir_size,
1764         .get_rxfh       = netvsc_get_rxfh,
1765         .set_rxfh       = netvsc_set_rxfh,
1766         .get_link_ksettings = netvsc_get_link_ksettings,
1767         .set_link_ksettings = netvsc_set_link_ksettings,
1768         .get_ringparam  = netvsc_get_ringparam,
1769         .set_ringparam  = netvsc_set_ringparam,
1770 };
1771
1772 static const struct net_device_ops device_ops = {
1773         .ndo_open =                     netvsc_open,
1774         .ndo_stop =                     netvsc_close,
1775         .ndo_start_xmit =               netvsc_start_xmit,
1776         .ndo_change_rx_flags =          netvsc_change_rx_flags,
1777         .ndo_set_rx_mode =              netvsc_set_rx_mode,
1778         .ndo_change_mtu =               netvsc_change_mtu,
1779         .ndo_validate_addr =            eth_validate_addr,
1780         .ndo_set_mac_address =          netvsc_set_mac_addr,
1781         .ndo_select_queue =             netvsc_select_queue,
1782         .ndo_get_stats64 =              netvsc_get_stats64,
1783 #ifdef CONFIG_NET_POLL_CONTROLLER
1784         .ndo_poll_controller =          netvsc_poll_controller,
1785 #endif
1786 };
1787
1788 /*
1789  * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
1790  * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
1791  * present send GARP packet to network peers with netif_notify_peers().
1792  */
1793 static void netvsc_link_change(struct work_struct *w)
1794 {
1795         struct net_device_context *ndev_ctx =
1796                 container_of(w, struct net_device_context, dwork.work);
1797         struct hv_device *device_obj = ndev_ctx->device_ctx;
1798         struct net_device *net = hv_get_drvdata(device_obj);
1799         struct netvsc_device *net_device;
1800         struct rndis_device *rdev;
1801         struct netvsc_reconfig *event = NULL;
1802         bool notify = false, reschedule = false;
1803         unsigned long flags, next_reconfig, delay;
1804
1805         /* if changes are happening, comeback later */
1806         if (!rtnl_trylock()) {
1807                 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1808                 return;
1809         }
1810
1811         net_device = rtnl_dereference(ndev_ctx->nvdev);
1812         if (!net_device)
1813                 goto out_unlock;
1814
1815         rdev = net_device->extension;
1816
1817         next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
1818         if (time_is_after_jiffies(next_reconfig)) {
1819                 /* link_watch only sends one notification with current state
1820                  * per second, avoid doing reconfig more frequently. Handle
1821                  * wrap around.
1822                  */
1823                 delay = next_reconfig - jiffies;
1824                 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
1825                 schedule_delayed_work(&ndev_ctx->dwork, delay);
1826                 goto out_unlock;
1827         }
1828         ndev_ctx->last_reconfig = jiffies;
1829
1830         spin_lock_irqsave(&ndev_ctx->lock, flags);
1831         if (!list_empty(&ndev_ctx->reconfig_events)) {
1832                 event = list_first_entry(&ndev_ctx->reconfig_events,
1833                                          struct netvsc_reconfig, list);
1834                 list_del(&event->list);
1835                 reschedule = !list_empty(&ndev_ctx->reconfig_events);
1836         }
1837         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1838
1839         if (!event)
1840                 goto out_unlock;
1841
1842         switch (event->event) {
1843                 /* Only the following events are possible due to the check in
1844                  * netvsc_linkstatus_callback()
1845                  */
1846         case RNDIS_STATUS_MEDIA_CONNECT:
1847                 if (rdev->link_state) {
1848                         rdev->link_state = false;
1849                         netif_carrier_on(net);
1850                         netif_tx_wake_all_queues(net);
1851                 } else {
1852                         notify = true;
1853                 }
1854                 kfree(event);
1855                 break;
1856         case RNDIS_STATUS_MEDIA_DISCONNECT:
1857                 if (!rdev->link_state) {
1858                         rdev->link_state = true;
1859                         netif_carrier_off(net);
1860                         netif_tx_stop_all_queues(net);
1861                 }
1862                 kfree(event);
1863                 break;
1864         case RNDIS_STATUS_NETWORK_CHANGE:
1865                 /* Only makes sense if carrier is present */
1866                 if (!rdev->link_state) {
1867                         rdev->link_state = true;
1868                         netif_carrier_off(net);
1869                         netif_tx_stop_all_queues(net);
1870                         event->event = RNDIS_STATUS_MEDIA_CONNECT;
1871                         spin_lock_irqsave(&ndev_ctx->lock, flags);
1872                         list_add(&event->list, &ndev_ctx->reconfig_events);
1873                         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1874                         reschedule = true;
1875                 }
1876                 break;
1877         }
1878
1879         rtnl_unlock();
1880
1881         if (notify)
1882                 netdev_notify_peers(net);
1883
1884         /* link_watch only sends one notification with current state per
1885          * second, handle next reconfig event in 2 seconds.
1886          */
1887         if (reschedule)
1888                 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1889
1890         return;
1891
1892 out_unlock:
1893         rtnl_unlock();
1894 }
1895
1896 static struct net_device *get_netvsc_bymac(const u8 *mac)
1897 {
1898         struct net_device_context *ndev_ctx;
1899
1900         list_for_each_entry(ndev_ctx, &netvsc_dev_list, list) {
1901                 struct net_device *dev = hv_get_drvdata(ndev_ctx->device_ctx);
1902
1903                 if (ether_addr_equal(mac, dev->perm_addr))
1904                         return dev;
1905         }
1906
1907         return NULL;
1908 }
1909
1910 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
1911 {
1912         struct net_device_context *net_device_ctx;
1913         struct net_device *dev;
1914
1915         dev = netdev_master_upper_dev_get(vf_netdev);
1916         if (!dev || dev->netdev_ops != &device_ops)
1917                 return NULL;    /* not a netvsc device */
1918
1919         net_device_ctx = netdev_priv(dev);
1920         if (!rtnl_dereference(net_device_ctx->nvdev))
1921                 return NULL;    /* device is removed */
1922
1923         return dev;
1924 }
1925
1926 /* Called when VF is injecting data into network stack.
1927  * Change the associated network device from VF to netvsc.
1928  * note: already called with rcu_read_lock
1929  */
1930 static rx_handler_result_t netvsc_vf_handle_frame(struct sk_buff **pskb)
1931 {
1932         struct sk_buff *skb = *pskb;
1933         struct net_device *ndev = rcu_dereference(skb->dev->rx_handler_data);
1934         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1935         struct netvsc_vf_pcpu_stats *pcpu_stats
1936                  = this_cpu_ptr(ndev_ctx->vf_stats);
1937
1938         skb->dev = ndev;
1939
1940         u64_stats_update_begin(&pcpu_stats->syncp);
1941         pcpu_stats->rx_packets++;
1942         pcpu_stats->rx_bytes += skb->len;
1943         u64_stats_update_end(&pcpu_stats->syncp);
1944
1945         return RX_HANDLER_ANOTHER;
1946 }
1947
1948 static int netvsc_vf_join(struct net_device *vf_netdev,
1949                           struct net_device *ndev)
1950 {
1951         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1952         int ret;
1953
1954         ret = netdev_rx_handler_register(vf_netdev,
1955                                          netvsc_vf_handle_frame, ndev);
1956         if (ret != 0) {
1957                 netdev_err(vf_netdev,
1958                            "can not register netvsc VF receive handler (err = %d)\n",
1959                            ret);
1960                 goto rx_handler_failed;
1961         }
1962
1963         ret = netdev_master_upper_dev_link(vf_netdev, ndev,
1964                                            NULL, NULL, NULL);
1965         if (ret != 0) {
1966                 netdev_err(vf_netdev,
1967                            "can not set master device %s (err = %d)\n",
1968                            ndev->name, ret);
1969                 goto upper_link_failed;
1970         }
1971
1972         /* set slave flag before open to prevent IPv6 addrconf */
1973         vf_netdev->flags |= IFF_SLAVE;
1974
1975         schedule_delayed_work(&ndev_ctx->vf_takeover, VF_TAKEOVER_INT);
1976
1977         call_netdevice_notifiers(NETDEV_JOIN, vf_netdev);
1978
1979         netdev_info(vf_netdev, "joined to %s\n", ndev->name);
1980         return 0;
1981
1982 upper_link_failed:
1983         netdev_rx_handler_unregister(vf_netdev);
1984 rx_handler_failed:
1985         return ret;
1986 }
1987
1988 static void __netvsc_vf_setup(struct net_device *ndev,
1989                               struct net_device *vf_netdev)
1990 {
1991         int ret;
1992
1993         /* Align MTU of VF with master */
1994         ret = dev_set_mtu(vf_netdev, ndev->mtu);
1995         if (ret)
1996                 netdev_warn(vf_netdev,
1997                             "unable to change mtu to %u\n", ndev->mtu);
1998
1999         /* set multicast etc flags on VF */
2000         dev_change_flags(vf_netdev, ndev->flags | IFF_SLAVE);
2001
2002         /* sync address list from ndev to VF */
2003         netif_addr_lock_bh(ndev);
2004         dev_uc_sync(vf_netdev, ndev);
2005         dev_mc_sync(vf_netdev, ndev);
2006         netif_addr_unlock_bh(ndev);
2007
2008         if (netif_running(ndev)) {
2009                 ret = dev_open(vf_netdev);
2010                 if (ret)
2011                         netdev_warn(vf_netdev,
2012                                     "unable to open: %d\n", ret);
2013         }
2014 }
2015
2016 /* Setup VF as slave of the synthetic device.
2017  * Runs in workqueue to avoid recursion in netlink callbacks.
2018  */
2019 static void netvsc_vf_setup(struct work_struct *w)
2020 {
2021         struct net_device_context *ndev_ctx
2022                 = container_of(w, struct net_device_context, vf_takeover.work);
2023         struct net_device *ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2024         struct net_device *vf_netdev;
2025
2026         if (!rtnl_trylock()) {
2027                 schedule_delayed_work(&ndev_ctx->vf_takeover, 0);
2028                 return;
2029         }
2030
2031         vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2032         if (vf_netdev)
2033                 __netvsc_vf_setup(ndev, vf_netdev);
2034
2035         rtnl_unlock();
2036 }
2037
2038 static int netvsc_register_vf(struct net_device *vf_netdev)
2039 {
2040         struct net_device *ndev;
2041         struct net_device_context *net_device_ctx;
2042         struct netvsc_device *netvsc_dev;
2043         int ret;
2044
2045         if (vf_netdev->addr_len != ETH_ALEN)
2046                 return NOTIFY_DONE;
2047
2048         /*
2049          * We will use the MAC address to locate the synthetic interface to
2050          * associate with the VF interface. If we don't find a matching
2051          * synthetic interface, move on.
2052          */
2053         ndev = get_netvsc_bymac(vf_netdev->perm_addr);
2054         if (!ndev)
2055                 return NOTIFY_DONE;
2056
2057         net_device_ctx = netdev_priv(ndev);
2058         netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2059         if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
2060                 return NOTIFY_DONE;
2061
2062         /* if syntihetic interface is a different namespace,
2063          * then move the VF to that namespace; join will be
2064          * done again in that context.
2065          */
2066         if (!net_eq(dev_net(ndev), dev_net(vf_netdev))) {
2067                 ret = dev_change_net_namespace(vf_netdev,
2068                                                dev_net(ndev), "eth%d");
2069                 if (ret)
2070                         netdev_err(vf_netdev,
2071                                    "could not move to same namespace as %s: %d\n",
2072                                    ndev->name, ret);
2073                 else
2074                         netdev_info(vf_netdev,
2075                                     "VF moved to namespace with: %s\n",
2076                                     ndev->name);
2077                 return NOTIFY_DONE;
2078         }
2079
2080         netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
2081
2082         if (netvsc_vf_join(vf_netdev, ndev) != 0)
2083                 return NOTIFY_DONE;
2084
2085         dev_hold(vf_netdev);
2086         rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
2087         return NOTIFY_OK;
2088 }
2089
2090 /* VF up/down change detected, schedule to change data path */
2091 static int netvsc_vf_changed(struct net_device *vf_netdev)
2092 {
2093         struct net_device_context *net_device_ctx;
2094         struct netvsc_device *netvsc_dev;
2095         struct net_device *ndev;
2096         bool vf_is_up = netif_running(vf_netdev);
2097
2098         ndev = get_netvsc_byref(vf_netdev);
2099         if (!ndev)
2100                 return NOTIFY_DONE;
2101
2102         net_device_ctx = netdev_priv(ndev);
2103         netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2104         if (!netvsc_dev)
2105                 return NOTIFY_DONE;
2106
2107         netvsc_switch_datapath(ndev, vf_is_up);
2108         netdev_info(ndev, "Data path switched %s VF: %s\n",
2109                     vf_is_up ? "to" : "from", vf_netdev->name);
2110
2111         return NOTIFY_OK;
2112 }
2113
2114 static int netvsc_unregister_vf(struct net_device *vf_netdev)
2115 {
2116         struct net_device *ndev;
2117         struct net_device_context *net_device_ctx;
2118
2119         ndev = get_netvsc_byref(vf_netdev);
2120         if (!ndev)
2121                 return NOTIFY_DONE;
2122
2123         net_device_ctx = netdev_priv(ndev);
2124         cancel_delayed_work_sync(&net_device_ctx->vf_takeover);
2125
2126         netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
2127
2128         netdev_rx_handler_unregister(vf_netdev);
2129         netdev_upper_dev_unlink(vf_netdev, ndev);
2130         RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
2131         dev_put(vf_netdev);
2132
2133         return NOTIFY_OK;
2134 }
2135
2136 static int netvsc_probe(struct hv_device *dev,
2137                         const struct hv_vmbus_device_id *dev_id)
2138 {
2139         struct net_device *net = NULL;
2140         struct net_device_context *net_device_ctx;
2141         struct netvsc_device_info device_info;
2142         struct netvsc_device *nvdev;
2143         int ret = -ENOMEM;
2144
2145         net = alloc_etherdev_mq(sizeof(struct net_device_context),
2146                                 VRSS_CHANNEL_MAX);
2147         if (!net)
2148                 goto no_net;
2149
2150         netif_carrier_off(net);
2151
2152         netvsc_init_settings(net);
2153
2154         net_device_ctx = netdev_priv(net);
2155         net_device_ctx->device_ctx = dev;
2156         net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
2157         if (netif_msg_probe(net_device_ctx))
2158                 netdev_dbg(net, "netvsc msg_enable: %d\n",
2159                            net_device_ctx->msg_enable);
2160
2161         hv_set_drvdata(dev, net);
2162
2163         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
2164
2165         spin_lock_init(&net_device_ctx->lock);
2166         INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
2167         INIT_DELAYED_WORK(&net_device_ctx->vf_takeover, netvsc_vf_setup);
2168
2169         net_device_ctx->vf_stats
2170                 = netdev_alloc_pcpu_stats(struct netvsc_vf_pcpu_stats);
2171         if (!net_device_ctx->vf_stats)
2172                 goto no_stats;
2173
2174         net->netdev_ops = &device_ops;
2175         net->ethtool_ops = &ethtool_ops;
2176         SET_NETDEV_DEV(net, &dev->device);
2177
2178         /* We always need headroom for rndis header */
2179         net->needed_headroom = RNDIS_AND_PPI_SIZE;
2180
2181         /* Initialize the number of queues to be 1, we may change it if more
2182          * channels are offered later.
2183          */
2184         netif_set_real_num_tx_queues(net, 1);
2185         netif_set_real_num_rx_queues(net, 1);
2186
2187         /* Notify the netvsc driver of the new device */
2188         memset(&device_info, 0, sizeof(device_info));
2189         device_info.num_chn = VRSS_CHANNEL_DEFAULT;
2190         device_info.send_sections = NETVSC_DEFAULT_TX;
2191         device_info.send_section_size = NETVSC_SEND_SECTION_SIZE;
2192         device_info.recv_sections = NETVSC_DEFAULT_RX;
2193         device_info.recv_section_size = NETVSC_RECV_SECTION_SIZE;
2194
2195         nvdev = rndis_filter_device_add(dev, &device_info);
2196         if (IS_ERR(nvdev)) {
2197                 ret = PTR_ERR(nvdev);
2198                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
2199                 goto rndis_failed;
2200         }
2201
2202         memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
2203
2204         if (nvdev->num_chn > 1)
2205                 schedule_work(&nvdev->subchan_work);
2206
2207         /* hw_features computed in rndis_netdev_set_hwcaps() */
2208         net->features = net->hw_features |
2209                 NETIF_F_HIGHDMA | NETIF_F_SG |
2210                 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
2211         net->vlan_features = net->features;
2212
2213         netdev_lockdep_set_classes(net);
2214
2215         /* MTU range: 68 - 1500 or 65521 */
2216         net->min_mtu = NETVSC_MTU_MIN;
2217         if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
2218                 net->max_mtu = NETVSC_MTU - ETH_HLEN;
2219         else
2220                 net->max_mtu = ETH_DATA_LEN;
2221
2222         rtnl_lock();
2223         ret = register_netdevice(net);
2224         if (ret != 0) {
2225                 pr_err("Unable to register netdev.\n");
2226                 goto register_failed;
2227         }
2228
2229         list_add(&net_device_ctx->list, &netvsc_dev_list);
2230         rtnl_unlock();
2231         return 0;
2232
2233 register_failed:
2234         rtnl_unlock();
2235         rndis_filter_device_remove(dev, nvdev);
2236 rndis_failed:
2237         free_percpu(net_device_ctx->vf_stats);
2238 no_stats:
2239         hv_set_drvdata(dev, NULL);
2240         free_netdev(net);
2241 no_net:
2242         return ret;
2243 }
2244
2245 static int netvsc_remove(struct hv_device *dev)
2246 {
2247         struct net_device_context *ndev_ctx;
2248         struct net_device *vf_netdev, *net;
2249         struct netvsc_device *nvdev;
2250
2251         net = hv_get_drvdata(dev);
2252         if (net == NULL) {
2253                 dev_err(&dev->device, "No net device to remove\n");
2254                 return 0;
2255         }
2256
2257         ndev_ctx = netdev_priv(net);
2258
2259         cancel_delayed_work_sync(&ndev_ctx->dwork);
2260
2261         rcu_read_lock();
2262         nvdev = rcu_dereference(ndev_ctx->nvdev);
2263
2264         if  (nvdev)
2265                 cancel_work_sync(&nvdev->subchan_work);
2266
2267         /*
2268          * Call to the vsc driver to let it know that the device is being
2269          * removed. Also blocks mtu and channel changes.
2270          */
2271         rtnl_lock();
2272         vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2273         if (vf_netdev)
2274                 netvsc_unregister_vf(vf_netdev);
2275
2276         if (nvdev)
2277                 rndis_filter_device_remove(dev, nvdev);
2278
2279         unregister_netdevice(net);
2280         list_del(&ndev_ctx->list);
2281
2282         rtnl_unlock();
2283         rcu_read_unlock();
2284
2285         hv_set_drvdata(dev, NULL);
2286
2287         free_percpu(ndev_ctx->vf_stats);
2288         free_netdev(net);
2289         return 0;
2290 }
2291
2292 static const struct hv_vmbus_device_id id_table[] = {
2293         /* Network guid */
2294         { HV_NIC_GUID, },
2295         { },
2296 };
2297
2298 MODULE_DEVICE_TABLE(vmbus, id_table);
2299
2300 /* The one and only one */
2301 static struct  hv_driver netvsc_drv = {
2302         .name = KBUILD_MODNAME,
2303         .id_table = id_table,
2304         .probe = netvsc_probe,
2305         .remove = netvsc_remove,
2306         .driver = {
2307                 .probe_type = PROBE_PREFER_ASYNCHRONOUS,
2308         },
2309 };
2310
2311 /*
2312  * On Hyper-V, every VF interface is matched with a corresponding
2313  * synthetic interface. The synthetic interface is presented first
2314  * to the guest. When the corresponding VF instance is registered,
2315  * we will take care of switching the data path.
2316  */
2317 static int netvsc_netdev_event(struct notifier_block *this,
2318                                unsigned long event, void *ptr)
2319 {
2320         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2321
2322         /* Skip our own events */
2323         if (event_dev->netdev_ops == &device_ops)
2324                 return NOTIFY_DONE;
2325
2326         /* Avoid non-Ethernet type devices */
2327         if (event_dev->type != ARPHRD_ETHER)
2328                 return NOTIFY_DONE;
2329
2330         /* Avoid Vlan dev with same MAC registering as VF */
2331         if (is_vlan_dev(event_dev))
2332                 return NOTIFY_DONE;
2333
2334         /* Avoid Bonding master dev with same MAC registering as VF */
2335         if ((event_dev->priv_flags & IFF_BONDING) &&
2336             (event_dev->flags & IFF_MASTER))
2337                 return NOTIFY_DONE;
2338
2339         switch (event) {
2340         case NETDEV_REGISTER:
2341                 return netvsc_register_vf(event_dev);
2342         case NETDEV_UNREGISTER:
2343                 return netvsc_unregister_vf(event_dev);
2344         case NETDEV_UP:
2345         case NETDEV_DOWN:
2346                 return netvsc_vf_changed(event_dev);
2347         default:
2348                 return NOTIFY_DONE;
2349         }
2350 }
2351
2352 static struct notifier_block netvsc_netdev_notifier = {
2353         .notifier_call = netvsc_netdev_event,
2354 };
2355
2356 static void __exit netvsc_drv_exit(void)
2357 {
2358         unregister_netdevice_notifier(&netvsc_netdev_notifier);
2359         vmbus_driver_unregister(&netvsc_drv);
2360 }
2361
2362 static int __init netvsc_drv_init(void)
2363 {
2364         int ret;
2365
2366         if (ring_size < RING_SIZE_MIN) {
2367                 ring_size = RING_SIZE_MIN;
2368                 pr_info("Increased ring_size to %u (min allowed)\n",
2369                         ring_size);
2370         }
2371         netvsc_ring_bytes = ring_size * PAGE_SIZE;
2372
2373         ret = vmbus_driver_register(&netvsc_drv);
2374         if (ret)
2375                 return ret;
2376
2377         register_netdevice_notifier(&netvsc_netdev_notifier);
2378         return 0;
2379 }
2380
2381 MODULE_LICENSE("GPL");
2382 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
2383
2384 module_init(netvsc_drv_init);
2385 module_exit(netvsc_drv_exit);