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 <haiyangz@microsoft.com>
  18  *   Hank Janssen  <hjanssen@microsoft.com>
  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 <net/arp.h>
  37 #include <net/route.h>
  38 #include <net/sock.h>
  39 #include <net/pkt_sched.h>
  40 #include <net/checksum.h>
  41 #include <net/ip6_checksum.h>
  42
  43 #include "hyperv_net.h"
  44
  45 #define RING_SIZE_MIN 64
  46 #define LINKCHANGE_INT (2 * HZ)
  47
  48 static int ring_size = 128;
  49 module_param(ring_size, int, S_IRUGO);
  50 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
  51
  52 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
  53                                 NETIF_MSG_LINK | NETIF_MSG_IFUP |
  54                                 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
  55                                 NETIF_MSG_TX_ERR;
  56
  57 static int debug = -1;
  58 module_param(debug, int, S_IRUGO);
  59 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  60
  61 static void netvsc_set_multicast_list(struct net_device *net)
  62 {
  63         struct net_device_context *net_device_ctx = netdev_priv(net);
  64         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
  65
  66         rndis_filter_update(nvdev);
  67 }
  68
  69 static int netvsc_open(struct net_device *net)
  70 {
  71         struct net_device_context *ndev_ctx = netdev_priv(net);
  72         struct netvsc_device *nvdev = ndev_ctx->nvdev;
  73         struct rndis_device *rdev;
  74         int ret = 0;
  75
  76         netif_carrier_off(net);
  77
  78         /* Open up the device */
  79         ret = rndis_filter_open(nvdev);
  80         if (ret != 0) {
  81                 netdev_err(net, "unable to open device (ret %d).\n", ret);
  82                 return ret;
  83         }
  84
  85         netif_tx_wake_all_queues(net);
  86
  87         rdev = nvdev->extension;
  88         if (!rdev->link_state && !ndev_ctx->datapath)
  89                 netif_carrier_on(net);
  90
  91         return ret;
  92 }
  93
  94 static int netvsc_close(struct net_device *net)
  95 {
  96         struct net_device_context *net_device_ctx = netdev_priv(net);
  97         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
  98         int ret;
  99         u32 aread, i, msec = 10, retry = 0, retry_max = 20;
 100         struct vmbus_channel *chn;
 101
 102         netif_tx_disable(net);
 103
 104         ret = rndis_filter_close(nvdev);
 105         if (ret != 0) {
 106                 netdev_err(net, "unable to close device (ret %d).\n", ret);
 107                 return ret;
 108         }
 109
 110         /* Ensure pending bytes in ring are read */
 111         while (true) {
 112                 aread = 0;
 113                 for (i = 0; i < nvdev->num_chn; i++) {
 114                         chn = nvdev->chan_table[i].channel;
 115                         if (!chn)
 116                                 continue;
 117
 118                         aread = hv_get_bytes_to_read(&chn->inbound);
 119                         if (aread)
 120                                 break;
 121
 122                         aread = hv_get_bytes_to_read(&chn->outbound);
 123                         if (aread)
 124                                 break;
 125                 }
 126
 127                 retry++;
 128                 if (retry > retry_max || aread == 0)
 129                         break;
 130
 131                 msleep(msec);
 132
 133                 if (msec < 1000)
 134                         msec *= 2;
 135         }
 136
 137         if (aread) {
 138                 netdev_err(net, "Ring buffer not empty after closing rndis\n");
 139                 ret = -ETIMEDOUT;
 140         }
 141
 142         return ret;
 143 }
 144
 145 static void *init_ppi_data(struct rndis_message *msg, u32 ppi_size,
 146                                 int pkt_type)
 147 {
 148         struct rndis_packet *rndis_pkt;
 149         struct rndis_per_packet_info *ppi;
 150
 151         rndis_pkt = &msg->msg.pkt;
 152         rndis_pkt->data_offset += ppi_size;
 153
 154         ppi = (struct rndis_per_packet_info *)((void *)rndis_pkt +
 155                 rndis_pkt->per_pkt_info_offset + rndis_pkt->per_pkt_info_len);
 156
 157         ppi->size = ppi_size;
 158         ppi->type = pkt_type;
 159         ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
 160
 161         rndis_pkt->per_pkt_info_len += ppi_size;
 162
 163         return ppi;
 164 }
 165
 166 /* Azure hosts don't support non-TCP port numbers in hashing yet. We compute
 167  * hash for non-TCP traffic with only IP numbers.
 168  */
 169 static inline u32 netvsc_get_hash(struct sk_buff *skb, struct sock *sk)
 170 {
 171         struct flow_keys flow;
 172         u32 hash;
 173         static u32 hashrnd __read_mostly;
 174
 175         net_get_random_once(&hashrnd, sizeof(hashrnd));
 176
 177         if (!skb_flow_dissect_flow_keys(skb, &flow, 0))
 178                 return 0;
 179
 180         if (flow.basic.ip_proto == IPPROTO_TCP) {
 181                 return skb_get_hash(skb);
 182         } else {
 183                 if (flow.basic.n_proto == htons(ETH_P_IP))
 184                         hash = jhash2((u32 *)&flow.addrs.v4addrs, 2, hashrnd);
 185                 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
 186                         hash = jhash2((u32 *)&flow.addrs.v6addrs, 8, hashrnd);
 187                 else
 188                         hash = 0;
 189
 190                 skb_set_hash(skb, hash, PKT_HASH_TYPE_L3);
 191         }
 192
 193         return hash;
 194 }
 195
 196 static inline int netvsc_get_tx_queue(struct net_device *ndev,
 197                                       struct sk_buff *skb, int old_idx)
 198 {
 199         const struct net_device_context *ndc = netdev_priv(ndev);
 200         struct sock *sk = skb->sk;
 201         int q_idx;
 202
 203         q_idx = ndc->tx_send_table[netvsc_get_hash(skb, sk) &
 204                                    (VRSS_SEND_TAB_SIZE - 1)];
 205
 206         /* If queue index changed record the new value */
 207         if (q_idx != old_idx &&
 208             sk && sk_fullsock(sk) && rcu_access_pointer(sk->sk_dst_cache))
 209                 sk_tx_queue_set(sk, q_idx);
 210
 211         return q_idx;
 212 }
 213
 214 /*
 215  * Select queue for transmit.
 216  *
 217  * If a valid queue has already been assigned, then use that.
 218  * Otherwise compute tx queue based on hash and the send table.
 219  *
 220  * This is basically similar to default (__netdev_pick_tx) with the added step
 221  * of using the host send_table when no other queue has been assigned.
 222  *
 223  * TODO support XPS - but get_xps_queue not exported
 224  */
 225 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
 226                         void *accel_priv, select_queue_fallback_t fallback)
 227 {
 228         unsigned int num_tx_queues = ndev->real_num_tx_queues;
 229         int q_idx = sk_tx_queue_get(skb->sk);
 230
 231         if (q_idx < 0 || skb->ooo_okay) {
 232                 /* If forwarding a packet, we use the recorded queue when
 233                  * available for better cache locality.
 234                  */
 235                 if (skb_rx_queue_recorded(skb))
 236                         q_idx = skb_get_rx_queue(skb);
 237                 else
 238                         q_idx = netvsc_get_tx_queue(ndev, skb, q_idx);
 239         }
 240
 241         while (unlikely(q_idx >= num_tx_queues))
 242                 q_idx -= num_tx_queues;
 243
 244         return q_idx;
 245 }
 246
 247 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
 248                         struct hv_page_buffer *pb)
 249 {
 250         int j = 0;
 251
 252         /* Deal with compund pages by ignoring unused part
 253          * of the page.
 254          */
 255         page += (offset >> PAGE_SHIFT);
 256         offset &= ~PAGE_MASK;
 257
 258         while (len > 0) {
 259                 unsigned long bytes;
 260
 261                 bytes = PAGE_SIZE - offset;
 262                 if (bytes > len)
 263                         bytes = len;
 264                 pb[j].pfn = page_to_pfn(page);
 265                 pb[j].offset = offset;
 266                 pb[j].len = bytes;
 267
 268                 offset += bytes;
 269                 len -= bytes;
 270
 271                 if (offset == PAGE_SIZE && len) {
 272                         page++;
 273                         offset = 0;
 274                         j++;
 275                 }
 276         }
 277
 278         return j + 1;
 279 }
 280
 281 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
 282                            struct hv_netvsc_packet *packet,
 283                            struct hv_page_buffer **page_buf)
 284 {
 285         struct hv_page_buffer *pb = *page_buf;
 286         u32 slots_used = 0;
 287         char *data = skb->data;
 288         int frags = skb_shinfo(skb)->nr_frags;
 289         int i;
 290
 291         /* The packet is laid out thus:
 292          * 1. hdr: RNDIS header and PPI
 293          * 2. skb linear data
 294          * 3. skb fragment data
 295          */
 296         if (hdr != NULL)
 297                 slots_used += fill_pg_buf(virt_to_page(hdr),
 298                                         offset_in_page(hdr),
 299                                         len, &pb[slots_used]);
 300
 301         packet->rmsg_size = len;
 302         packet->rmsg_pgcnt = slots_used;
 303
 304         slots_used += fill_pg_buf(virt_to_page(data),
 305                                 offset_in_page(data),
 306                                 skb_headlen(skb), &pb[slots_used]);
 307
 308         for (i = 0; i < frags; i++) {
 309                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 310
 311                 slots_used += fill_pg_buf(skb_frag_page(frag),
 312                                         frag->page_offset,
 313                                         skb_frag_size(frag), &pb[slots_used]);
 314         }
 315         return slots_used;
 316 }
 317
 318 /* Estimate number of page buffers neede to transmit
 319  * Need at most 2 for RNDIS header plus skb body and fragments.
 320  */
 321 static unsigned int netvsc_get_slots(const struct sk_buff *skb)
 322 {
 323         return PFN_UP(offset_in_page(skb->data) + skb_headlen(skb))
 324                 + skb_shinfo(skb)->nr_frags
 325                 + 2;
 326 }
 327
 328 static u32 net_checksum_info(struct sk_buff *skb)
 329 {
 330         if (skb->protocol == htons(ETH_P_IP)) {
 331                 struct iphdr *ip = ip_hdr(skb);
 332
 333                 if (ip->protocol == IPPROTO_TCP)
 334                         return TRANSPORT_INFO_IPV4_TCP;
 335                 else if (ip->protocol == IPPROTO_UDP)
 336                         return TRANSPORT_INFO_IPV4_UDP;
 337         } else {
 338                 struct ipv6hdr *ip6 = ipv6_hdr(skb);
 339
 340                 if (ip6->nexthdr == IPPROTO_TCP)
 341                         return TRANSPORT_INFO_IPV6_TCP;
 342                 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
 343                         return TRANSPORT_INFO_IPV6_UDP;
 344         }
 345
 346         return TRANSPORT_INFO_NOT_IP;
 347 }
 348
 349 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
 350 {
 351         struct net_device_context *net_device_ctx = netdev_priv(net);
 352         struct hv_netvsc_packet *packet = NULL;
 353         int ret;
 354         unsigned int num_data_pgs;
 355         struct rndis_message *rndis_msg;
 356         struct rndis_packet *rndis_pkt;
 357         u32 rndis_msg_size;
 358         struct rndis_per_packet_info *ppi;
 359         u32 hash;
 360         struct hv_page_buffer page_buf[MAX_PAGE_BUFFER_COUNT];
 361         struct hv_page_buffer *pb = page_buf;
 362
 363         /* We can only transmit MAX_PAGE_BUFFER_COUNT number
 364          * of pages in a single packet. If skb is scattered around
 365          * more pages we try linearizing it.
 366          */
 367         num_data_pgs = netvsc_get_slots(skb);
 368         if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
 369                 ++net_device_ctx->eth_stats.tx_scattered;
 370
 371                 if (skb_linearize(skb))
 372                         goto no_memory;
 373
 374                 num_data_pgs = netvsc_get_slots(skb);
 375                 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
 376                         ++net_device_ctx->eth_stats.tx_too_big;
 377                         goto drop;
 378                 }
 379         }
 380
 381         /*
 382          * Place the rndis header in the skb head room and
 383          * the skb->cb will be used for hv_netvsc_packet
 384          * structure.
 385          */
 386         ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
 387         if (ret)
 388                 goto no_memory;
 389
 390         /* Use the skb control buffer for building up the packet */
 391         BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
 392                         FIELD_SIZEOF(struct sk_buff, cb));
 393         packet = (struct hv_netvsc_packet *)skb->cb;
 394
 395         packet->q_idx = skb_get_queue_mapping(skb);
 396
 397         packet->total_data_buflen = skb->len;
 398         packet->total_bytes = skb->len;
 399         packet->total_packets = 1;
 400
 401         rndis_msg = (struct rndis_message *)skb->head;
 402
 403         memset(rndis_msg, 0, RNDIS_AND_PPI_SIZE);
 404
 405         /* Add the rndis header */
 406         rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
 407         rndis_msg->msg_len = packet->total_data_buflen;
 408         rndis_pkt = &rndis_msg->msg.pkt;
 409         rndis_pkt->data_offset = sizeof(struct rndis_packet);
 410         rndis_pkt->data_len = packet->total_data_buflen;
 411         rndis_pkt->per_pkt_info_offset = sizeof(struct rndis_packet);
 412
 413         rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
 414
 415         hash = skb_get_hash_raw(skb);
 416         if (hash != 0 && net->real_num_tx_queues > 1) {
 417                 rndis_msg_size += NDIS_HASH_PPI_SIZE;
 418                 ppi = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
 419                                     NBL_HASH_VALUE);
 420                 *(u32 *)((void *)ppi + ppi->ppi_offset) = hash;
 421         }
 422
 423         if (skb_vlan_tag_present(skb)) {
 424                 struct ndis_pkt_8021q_info *vlan;
 425
 426                 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
 427                 ppi = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
 428                                         IEEE_8021Q_INFO);
 429                 vlan = (struct ndis_pkt_8021q_info *)((void *)ppi +
 430                                                 ppi->ppi_offset);
 431                 vlan->vlanid = skb->vlan_tci & VLAN_VID_MASK;
 432                 vlan->pri = (skb->vlan_tci & VLAN_PRIO_MASK) >>
 433                                 VLAN_PRIO_SHIFT;
 434         }
 435
 436         if (skb_is_gso(skb)) {
 437                 struct ndis_tcp_lso_info *lso_info;
 438
 439                 rndis_msg_size += NDIS_LSO_PPI_SIZE;
 440                 ppi = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
 441                                     TCP_LARGESEND_PKTINFO);
 442
 443                 lso_info = (struct ndis_tcp_lso_info *)((void *)ppi +
 444                                                         ppi->ppi_offset);
 445
 446                 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
 447                 if (skb->protocol == htons(ETH_P_IP)) {
 448                         lso_info->lso_v2_transmit.ip_version =
 449                                 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
 450                         ip_hdr(skb)->tot_len = 0;
 451                         ip_hdr(skb)->check = 0;
 452                         tcp_hdr(skb)->check =
 453                                 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
 454                                                    ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 455                 } else {
 456                         lso_info->lso_v2_transmit.ip_version =
 457                                 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
 458                         ipv6_hdr(skb)->payload_len = 0;
 459                         tcp_hdr(skb)->check =
 460                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 461                                                  &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 462                 }
 463                 lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb);
 464                 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
 465         } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
 466                 if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) {
 467                         struct ndis_tcp_ip_checksum_info *csum_info;
 468
 469                         rndis_msg_size += NDIS_CSUM_PPI_SIZE;
 470                         ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
 471                                             TCPIP_CHKSUM_PKTINFO);
 472
 473                         csum_info = (struct ndis_tcp_ip_checksum_info *)((void *)ppi +
 474                                                                          ppi->ppi_offset);
 475
 476                         csum_info->transmit.tcp_header_offset = skb_transport_offset(skb);
 477
 478                         if (skb->protocol == htons(ETH_P_IP)) {
 479                                 csum_info->transmit.is_ipv4 = 1;
 480
 481                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
 482                                         csum_info->transmit.tcp_checksum = 1;
 483                                 else
 484                                         csum_info->transmit.udp_checksum = 1;
 485                         } else {
 486                                 csum_info->transmit.is_ipv6 = 1;
 487
 488                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
 489                                         csum_info->transmit.tcp_checksum = 1;
 490                                 else
 491                                         csum_info->transmit.udp_checksum = 1;
 492                         }
 493                 } else {
 494                         /* Can't do offload of this type of checksum */
 495                         if (skb_checksum_help(skb))
 496                                 goto drop;
 497                 }
 498         }
 499
 500         /* Start filling in the page buffers with the rndis hdr */
 501         rndis_msg->msg_len += rndis_msg_size;
 502         packet->total_data_buflen = rndis_msg->msg_len;
 503         packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
 504                                                skb, packet, &pb);
 505
 506         /* timestamp packet in software */
 507         skb_tx_timestamp(skb);
 508         ret = netvsc_send(net_device_ctx->device_ctx, packet,
 509                           rndis_msg, &pb, skb);
 510         if (likely(ret == 0))
 511                 return NETDEV_TX_OK;
 512
 513         if (ret == -EAGAIN) {
 514                 ++net_device_ctx->eth_stats.tx_busy;
 515                 return NETDEV_TX_BUSY;
 516         }
 517
 518         if (ret == -ENOSPC)
 519                 ++net_device_ctx->eth_stats.tx_no_space;
 520
 521 drop:
 522         dev_kfree_skb_any(skb);
 523         net->stats.tx_dropped++;
 524
 525         return NETDEV_TX_OK;
 526
 527 no_memory:
 528         ++net_device_ctx->eth_stats.tx_no_memory;
 529         goto drop;
 530 }
 531 /*
 532  * netvsc_linkstatus_callback - Link up/down notification
 533  */
 534 void netvsc_linkstatus_callback(struct hv_device *device_obj,
 535                                 struct rndis_message *resp)
 536 {
 537         struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
 538         struct net_device *net;
 539         struct net_device_context *ndev_ctx;
 540         struct netvsc_reconfig *event;
 541         unsigned long flags;
 542
 543         net = hv_get_drvdata(device_obj);
 544
 545         if (!net)
 546                 return;
 547
 548         ndev_ctx = netdev_priv(net);
 549
 550         /* Update the physical link speed when changing to another vSwitch */
 551         if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
 552                 u32 speed;
 553
 554                 speed = *(u32 *)((void *)indicate + indicate->
 555                                  status_buf_offset) / 10000;
 556                 ndev_ctx->speed = speed;
 557                 return;
 558         }
 559
 560         /* Handle these link change statuses below */
 561         if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
 562             indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
 563             indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
 564                 return;
 565
 566         if (net->reg_state != NETREG_REGISTERED)
 567                 return;
 568
 569         event = kzalloc(sizeof(*event), GFP_ATOMIC);
 570         if (!event)
 571                 return;
 572         event->event = indicate->status;
 573
 574         spin_lock_irqsave(&ndev_ctx->lock, flags);
 575         list_add_tail(&event->list, &ndev_ctx->reconfig_events);
 576         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
 577
 578         schedule_delayed_work(&ndev_ctx->dwork, 0);
 579 }
 580
 581 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
 582                                              struct napi_struct *napi,
 583                                              const struct ndis_tcp_ip_checksum_info *csum_info,
 584                                              const struct ndis_pkt_8021q_info *vlan,
 585                                              void *data, u32 buflen)
 586 {
 587         struct sk_buff *skb;
 588
 589         skb = napi_alloc_skb(napi, buflen);
 590         if (!skb)
 591                 return skb;
 592
 593         /*
 594          * Copy to skb. This copy is needed here since the memory pointed by
 595          * hv_netvsc_packet cannot be deallocated
 596          */
 597         skb_put_data(skb, data, buflen);
 598
 599         skb->protocol = eth_type_trans(skb, net);
 600
 601         /* skb is already created with CHECKSUM_NONE */
 602         skb_checksum_none_assert(skb);
 603
 604         /*
 605          * In Linux, the IP checksum is always checked.
 606          * Do L4 checksum offload if enabled and present.
 607          */
 608         if (csum_info && (net->features & NETIF_F_RXCSUM)) {
 609                 if (csum_info->receive.tcp_checksum_succeeded ||
 610                     csum_info->receive.udp_checksum_succeeded)
 611                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 612         }
 613
 614         if (vlan) {
 615                 u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT);
 616
 617                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 618                                        vlan_tci);
 619         }
 620
 621         return skb;
 622 }
 623
 624 /*
 625  * netvsc_recv_callback -  Callback when we receive a packet from the
 626  * "wire" on the specified device.
 627  */
 628 int netvsc_recv_callback(struct net_device *net,
 629                          struct vmbus_channel *channel,
 630                          void  *data, u32 len,
 631                          const struct ndis_tcp_ip_checksum_info *csum_info,
 632                          const struct ndis_pkt_8021q_info *vlan)
 633 {
 634         struct net_device_context *net_device_ctx = netdev_priv(net);
 635         struct netvsc_device *net_device;
 636         u16 q_idx = channel->offermsg.offer.sub_channel_index;
 637         struct netvsc_channel *nvchan;
 638         struct net_device *vf_netdev;
 639         struct sk_buff *skb;
 640         struct netvsc_stats *rx_stats;
 641
 642         if (net->reg_state != NETREG_REGISTERED)
 643                 return NVSP_STAT_FAIL;
 644
 645         /*
 646          * If necessary, inject this packet into the VF interface.
 647          * On Hyper-V, multicast and brodcast packets are only delivered
 648          * to the synthetic interface (after subjecting these to
 649          * policy filters on the host). Deliver these via the VF
 650          * interface in the guest.
 651          */
 652         rcu_read_lock();
 653         net_device = rcu_dereference(net_device_ctx->nvdev);
 654         if (unlikely(!net_device))
 655                 goto drop;
 656
 657         nvchan = &net_device->chan_table[q_idx];
 658         vf_netdev = rcu_dereference(net_device_ctx->vf_netdev);
 659         if (vf_netdev && (vf_netdev->flags & IFF_UP))
 660                 net = vf_netdev;
 661
 662         /* Allocate a skb - TODO direct I/O to pages? */
 663         skb = netvsc_alloc_recv_skb(net, &nvchan->napi,
 664                                     csum_info, vlan, data, len);
 665         if (unlikely(!skb)) {
 666 drop:
 667                 ++net->stats.rx_dropped;
 668                 rcu_read_unlock();
 669                 return NVSP_STAT_FAIL;
 670         }
 671
 672         if (net != vf_netdev)
 673                 skb_record_rx_queue(skb, q_idx);
 674
 675         /*
 676          * Even if injecting the packet, record the statistics
 677          * on the synthetic device because modifying the VF device
 678          * statistics will not work correctly.
 679          */
 680         rx_stats = &nvchan->rx_stats;
 681         u64_stats_update_begin(&rx_stats->syncp);
 682         rx_stats->packets++;
 683         rx_stats->bytes += len;
 684
 685         if (skb->pkt_type == PACKET_BROADCAST)
 686                 ++rx_stats->broadcast;
 687         else if (skb->pkt_type == PACKET_MULTICAST)
 688                 ++rx_stats->multicast;
 689         u64_stats_update_end(&rx_stats->syncp);
 690
 691         napi_gro_receive(&nvchan->napi, skb);
 692         rcu_read_unlock();
 693
 694         return 0;
 695 }
 696
 697 static void netvsc_get_drvinfo(struct net_device *net,
 698                                struct ethtool_drvinfo *info)
 699 {
 700         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
 701         strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
 702 }
 703
 704 static void netvsc_get_channels(struct net_device *net,
 705                                 struct ethtool_channels *channel)
 706 {
 707         struct net_device_context *net_device_ctx = netdev_priv(net);
 708         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
 709
 710         if (nvdev) {
 711                 channel->max_combined   = nvdev->max_chn;
 712                 channel->combined_count = nvdev->num_chn;
 713         }
 714 }
 715
 716 static int netvsc_set_queues(struct net_device *net, struct hv_device *dev,
 717                              u32 num_chn)
 718 {
 719         struct netvsc_device_info device_info;
 720         int ret;
 721
 722         memset(&device_info, 0, sizeof(device_info));
 723         device_info.num_chn = num_chn;
 724         device_info.ring_size = ring_size;
 725         device_info.max_num_vrss_chns = num_chn;
 726
 727         ret = rndis_filter_device_add(dev, &device_info);
 728         if (ret)
 729                 return ret;
 730
 731         ret = netif_set_real_num_tx_queues(net, num_chn);
 732         if (ret)
 733                 return ret;
 734
 735         ret = netif_set_real_num_rx_queues(net, num_chn);
 736
 737         return ret;
 738 }
 739
 740 static int netvsc_set_channels(struct net_device *net,
 741                                struct ethtool_channels *channels)
 742 {
 743         struct net_device_context *net_device_ctx = netdev_priv(net);
 744         struct hv_device *dev = net_device_ctx->device_ctx;
 745         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
 746         unsigned int count = channels->combined_count;
 747         bool was_running;
 748         int ret;
 749
 750         /* We do not support separate count for rx, tx, or other */
 751         if (count == 0 ||
 752             channels->rx_count || channels->tx_count || channels->other_count)
 753                 return -EINVAL;
 754
 755         if (count > net->num_tx_queues || count > VRSS_CHANNEL_MAX)
 756                 return -EINVAL;
 757
 758         if (!nvdev || nvdev->destroy)
 759                 return -ENODEV;
 760
 761         if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
 762                 return -EINVAL;
 763
 764         if (count > nvdev->max_chn)
 765                 return -EINVAL;
 766
 767         was_running = netif_running(net);
 768         if (was_running) {
 769                 ret = netvsc_close(net);
 770                 if (ret)
 771                         return ret;
 772         }
 773
 774         rndis_filter_device_remove(dev, nvdev);
 775
 776         ret = netvsc_set_queues(net, dev, count);
 777         if (ret == 0)
 778                 nvdev->num_chn = count;
 779         else
 780                 netvsc_set_queues(net, dev, nvdev->num_chn);
 781
 782         if (was_running)
 783                 ret = netvsc_open(net);
 784
 785         /* We may have missed link change notifications */
 786         schedule_delayed_work(&net_device_ctx->dwork, 0);
 787
 788         return ret;
 789 }
 790
 791 static bool
 792 netvsc_validate_ethtool_ss_cmd(const struct ethtool_link_ksettings *cmd)
 793 {
 794         struct ethtool_link_ksettings diff1 = *cmd;
 795         struct ethtool_link_ksettings diff2 = {};
 796
 797         diff1.base.speed = 0;
 798         diff1.base.duplex = 0;
 799         /* advertising and cmd are usually set */
 800         ethtool_link_ksettings_zero_link_mode(&diff1, advertising);
 801         diff1.base.cmd = 0;
 802         /* We set port to PORT_OTHER */
 803         diff2.base.port = PORT_OTHER;
 804
 805         return !memcmp(&diff1, &diff2, sizeof(diff1));
 806 }
 807
 808 static void netvsc_init_settings(struct net_device *dev)
 809 {
 810         struct net_device_context *ndc = netdev_priv(dev);
 811
 812         ndc->speed = SPEED_UNKNOWN;
 813         ndc->duplex = DUPLEX_FULL;
 814 }
 815
 816 static int netvsc_get_link_ksettings(struct net_device *dev,
 817                                      struct ethtool_link_ksettings *cmd)
 818 {
 819         struct net_device_context *ndc = netdev_priv(dev);
 820
 821         cmd->base.speed = ndc->speed;
 822         cmd->base.duplex = ndc->duplex;
 823         cmd->base.port = PORT_OTHER;
 824
 825         return 0;
 826 }
 827
 828 static int netvsc_set_link_ksettings(struct net_device *dev,
 829                                      const struct ethtool_link_ksettings *cmd)
 830 {
 831         struct net_device_context *ndc = netdev_priv(dev);
 832         u32 speed;
 833
 834         speed = cmd->base.speed;
 835         if (!ethtool_validate_speed(speed) ||
 836             !ethtool_validate_duplex(cmd->base.duplex) ||
 837             !netvsc_validate_ethtool_ss_cmd(cmd))
 838                 return -EINVAL;
 839
 840         ndc->speed = speed;
 841         ndc->duplex = cmd->base.duplex;
 842
 843         return 0;
 844 }
 845
 846 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
 847 {
 848         struct net_device_context *ndevctx = netdev_priv(ndev);
 849         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
 850         struct hv_device *hdev = ndevctx->device_ctx;
 851         struct netvsc_device_info device_info;
 852         bool was_running;
 853         int ret = 0;
 854
 855         if (!nvdev || nvdev->destroy)
 856                 return -ENODEV;
 857
 858         was_running = netif_running(ndev);
 859         if (was_running) {
 860                 ret = netvsc_close(ndev);
 861                 if (ret)
 862                         return ret;
 863         }
 864
 865         memset(&device_info, 0, sizeof(device_info));
 866         device_info.ring_size = ring_size;
 867         device_info.num_chn = nvdev->num_chn;
 868         device_info.max_num_vrss_chns = nvdev->num_chn;
 869
 870         rndis_filter_device_remove(hdev, nvdev);
 871
 872         /* 'nvdev' has been freed in rndis_filter_device_remove() ->
 873          * netvsc_device_remove () -> free_netvsc_device().
 874          * We mustn't access it before it's re-created in
 875          * rndis_filter_device_add() -> netvsc_device_add().
 876          */
 877
 878         ndev->mtu = mtu;
 879
 880         rndis_filter_device_add(hdev, &device_info);
 881
 882         if (was_running)
 883                 ret = netvsc_open(ndev);
 884
 885         /* We may have missed link change notifications */
 886         schedule_delayed_work(&ndevctx->dwork, 0);
 887
 888         return ret;
 889 }
 890
 891 static void netvsc_get_stats64(struct net_device *net,
 892                                struct rtnl_link_stats64 *t)
 893 {
 894         struct net_device_context *ndev_ctx = netdev_priv(net);
 895         struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
 896         int i;
 897
 898         if (!nvdev)
 899                 return;
 900
 901         for (i = 0; i < nvdev->num_chn; i++) {
 902                 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
 903                 const struct netvsc_stats *stats;
 904                 u64 packets, bytes, multicast;
 905                 unsigned int start;
 906
 907                 stats = &nvchan->tx_stats;
 908                 do {
 909                         start = u64_stats_fetch_begin_irq(&stats->syncp);
 910                         packets = stats->packets;
 911                         bytes = stats->bytes;
 912                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
 913
 914                 t->tx_bytes     += bytes;
 915                 t->tx_packets   += packets;
 916
 917                 stats = &nvchan->rx_stats;
 918                 do {
 919                         start = u64_stats_fetch_begin_irq(&stats->syncp);
 920                         packets = stats->packets;
 921                         bytes = stats->bytes;
 922                         multicast = stats->multicast + stats->broadcast;
 923                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
 924
 925                 t->rx_bytes     += bytes;
 926                 t->rx_packets   += packets;
 927                 t->multicast    += multicast;
 928         }
 929
 930         t->tx_dropped   = net->stats.tx_dropped;
 931         t->tx_errors    = net->stats.tx_errors;
 932
 933         t->rx_dropped   = net->stats.rx_dropped;
 934         t->rx_errors    = net->stats.rx_errors;
 935 }
 936
 937 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
 938 {
 939         struct sockaddr *addr = p;
 940         char save_adr[ETH_ALEN];
 941         unsigned char save_aatype;
 942         int err;
 943
 944         memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
 945         save_aatype = ndev->addr_assign_type;
 946
 947         err = eth_mac_addr(ndev, p);
 948         if (err != 0)
 949                 return err;
 950
 951         err = rndis_filter_set_device_mac(ndev, addr->sa_data);
 952         if (err != 0) {
 953                 /* roll back to saved MAC */
 954                 memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
 955                 ndev->addr_assign_type = save_aatype;
 956         }
 957
 958         return err;
 959 }
 960
 961 static const struct {
 962         char name[ETH_GSTRING_LEN];
 963         u16 offset;
 964 } netvsc_stats[] = {
 965         { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
 966         { "tx_no_memory",  offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
 967         { "tx_no_space",  offsetof(struct netvsc_ethtool_stats, tx_no_space) },
 968         { "tx_too_big",   offsetof(struct netvsc_ethtool_stats, tx_too_big) },
 969         { "tx_busy",      offsetof(struct netvsc_ethtool_stats, tx_busy) },
 970 };
 971
 972 #define NETVSC_GLOBAL_STATS_LEN ARRAY_SIZE(netvsc_stats)
 973
 974 /* 4 statistics per queue (rx/tx packets/bytes) */
 975 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 4)
 976
 977 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
 978 {
 979         struct net_device_context *ndc = netdev_priv(dev);
 980         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
 981
 982         if (!nvdev)
 983                 return -ENODEV;
 984
 985         switch (string_set) {
 986         case ETH_SS_STATS:
 987                 return NETVSC_GLOBAL_STATS_LEN + NETVSC_QUEUE_STATS_LEN(nvdev);
 988         default:
 989                 return -EINVAL;
 990         }
 991 }
 992
 993 static void netvsc_get_ethtool_stats(struct net_device *dev,
 994                                      struct ethtool_stats *stats, u64 *data)
 995 {
 996         struct net_device_context *ndc = netdev_priv(dev);
 997         struct netvsc_device *nvdev = rcu_dereference(ndc->nvdev);
 998         const void *nds = &ndc->eth_stats;
 999         const struct netvsc_stats *qstats;
1000         unsigned int start;
1001         u64 packets, bytes;
1002         int i, j;
1003
1004         if (!nvdev)
1005                 return;
1006
1007         for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
1008                 data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1009
1010         for (j = 0; j < nvdev->num_chn; j++) {
1011                 qstats = &nvdev->chan_table[j].tx_stats;
1012
1013                 do {
1014                         start = u64_stats_fetch_begin_irq(&qstats->syncp);
1015                         packets = qstats->packets;
1016                         bytes = qstats->bytes;
1017                 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1018                 data[i++] = packets;
1019                 data[i++] = bytes;
1020
1021                 qstats = &nvdev->chan_table[j].rx_stats;
1022                 do {
1023                         start = u64_stats_fetch_begin_irq(&qstats->syncp);
1024                         packets = qstats->packets;
1025                         bytes = qstats->bytes;
1026                 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1027                 data[i++] = packets;
1028                 data[i++] = bytes;
1029         }
1030 }
1031
1032 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1033 {
1034         struct net_device_context *ndc = netdev_priv(dev);
1035         struct netvsc_device *nvdev = rcu_dereference(ndc->nvdev);
1036         u8 *p = data;
1037         int i;
1038
1039         if (!nvdev)
1040                 return;
1041
1042         switch (stringset) {
1043         case ETH_SS_STATS:
1044                 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++)
1045                         memcpy(p + i * ETH_GSTRING_LEN,
1046                                netvsc_stats[i].name, ETH_GSTRING_LEN);
1047
1048                 p += i * ETH_GSTRING_LEN;
1049                 for (i = 0; i < nvdev->num_chn; i++) {
1050                         sprintf(p, "tx_queue_%u_packets", i);
1051                         p += ETH_GSTRING_LEN;
1052                         sprintf(p, "tx_queue_%u_bytes", i);
1053                         p += ETH_GSTRING_LEN;
1054                         sprintf(p, "rx_queue_%u_packets", i);
1055                         p += ETH_GSTRING_LEN;
1056                         sprintf(p, "rx_queue_%u_bytes", i);
1057                         p += ETH_GSTRING_LEN;
1058                 }
1059
1060                 break;
1061         }
1062 }
1063
1064 static int
1065 netvsc_get_rss_hash_opts(struct netvsc_device *nvdev,
1066                          struct ethtool_rxnfc *info)
1067 {
1068         info->data = RXH_IP_SRC | RXH_IP_DST;
1069
1070         switch (info->flow_type) {
1071         case TCP_V4_FLOW:
1072         case TCP_V6_FLOW:
1073                 info->data |= RXH_L4_B_0_1 | RXH_L4_B_2_3;
1074                 /* fallthrough */
1075         case UDP_V4_FLOW:
1076         case UDP_V6_FLOW:
1077         case IPV4_FLOW:
1078         case IPV6_FLOW:
1079                 break;
1080         default:
1081                 info->data = 0;
1082                 break;
1083         }
1084
1085         return 0;
1086 }
1087
1088 static int
1089 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1090                  u32 *rules)
1091 {
1092         struct net_device_context *ndc = netdev_priv(dev);
1093         struct netvsc_device *nvdev = rcu_dereference(ndc->nvdev);
1094
1095         if (!nvdev)
1096                 return -ENODEV;
1097
1098         switch (info->cmd) {
1099         case ETHTOOL_GRXRINGS:
1100                 info->data = nvdev->num_chn;
1101                 return 0;
1102
1103         case ETHTOOL_GRXFH:
1104                 return netvsc_get_rss_hash_opts(nvdev, info);
1105         }
1106         return -EOPNOTSUPP;
1107 }
1108
1109 #ifdef CONFIG_NET_POLL_CONTROLLER
1110 static void netvsc_poll_controller(struct net_device *dev)
1111 {
1112         struct net_device_context *ndc = netdev_priv(dev);
1113         struct netvsc_device *ndev;
1114         int i;
1115
1116         rcu_read_lock();
1117         ndev = rcu_dereference(ndc->nvdev);
1118         if (ndev) {
1119                 for (i = 0; i < ndev->num_chn; i++) {
1120                         struct netvsc_channel *nvchan = &ndev->chan_table[i];
1121
1122                         napi_schedule(&nvchan->napi);
1123                 }
1124         }
1125         rcu_read_unlock();
1126 }
1127 #endif
1128
1129 static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1130 {
1131         return NETVSC_HASH_KEYLEN;
1132 }
1133
1134 static u32 netvsc_rss_indir_size(struct net_device *dev)
1135 {
1136         return ITAB_NUM;
1137 }
1138
1139 static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key,
1140                            u8 *hfunc)
1141 {
1142         struct net_device_context *ndc = netdev_priv(dev);
1143         struct netvsc_device *ndev = rcu_dereference(ndc->nvdev);
1144         struct rndis_device *rndis_dev;
1145         int i;
1146
1147         if (!ndev)
1148                 return -ENODEV;
1149
1150         if (hfunc)
1151                 *hfunc = ETH_RSS_HASH_TOP;      /* Toeplitz */
1152
1153         rndis_dev = ndev->extension;
1154         if (indir) {
1155                 for (i = 0; i < ITAB_NUM; i++)
1156                         indir[i] = rndis_dev->ind_table[i];
1157         }
1158
1159         if (key)
1160                 memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1161
1162         return 0;
1163 }
1164
1165 static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
1166                            const u8 *key, const u8 hfunc)
1167 {
1168         struct net_device_context *ndc = netdev_priv(dev);
1169         struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1170         struct rndis_device *rndis_dev;
1171         int i;
1172
1173         if (!ndev)
1174                 return -ENODEV;
1175
1176         if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
1177                 return -EOPNOTSUPP;
1178
1179         rndis_dev = ndev->extension;
1180         if (indir) {
1181                 for (i = 0; i < ITAB_NUM; i++)
1182                         if (indir[i] >= VRSS_CHANNEL_MAX)
1183                                 return -EINVAL;
1184
1185                 for (i = 0; i < ITAB_NUM; i++)
1186                         rndis_dev->ind_table[i] = indir[i];
1187         }
1188
1189         if (!key) {
1190                 if (!indir)
1191                         return 0;
1192
1193                 key = rndis_dev->rss_key;
1194         }
1195
1196         return rndis_filter_set_rss_param(rndis_dev, key, ndev->num_chn);
1197 }
1198
1199 static const struct ethtool_ops ethtool_ops = {
1200         .get_drvinfo    = netvsc_get_drvinfo,
1201         .get_link       = ethtool_op_get_link,
1202         .get_ethtool_stats = netvsc_get_ethtool_stats,
1203         .get_sset_count = netvsc_get_sset_count,
1204         .get_strings    = netvsc_get_strings,
1205         .get_channels   = netvsc_get_channels,
1206         .set_channels   = netvsc_set_channels,
1207         .get_ts_info    = ethtool_op_get_ts_info,
1208         .get_rxnfc      = netvsc_get_rxnfc,
1209         .get_rxfh_key_size = netvsc_get_rxfh_key_size,
1210         .get_rxfh_indir_size = netvsc_rss_indir_size,
1211         .get_rxfh       = netvsc_get_rxfh,
1212         .set_rxfh       = netvsc_set_rxfh,
1213         .get_link_ksettings = netvsc_get_link_ksettings,
1214         .set_link_ksettings = netvsc_set_link_ksettings,
1215 };
1216
1217 static const struct net_device_ops device_ops = {
1218         .ndo_open =                     netvsc_open,
1219         .ndo_stop =                     netvsc_close,
1220         .ndo_start_xmit =               netvsc_start_xmit,
1221         .ndo_set_rx_mode =              netvsc_set_multicast_list,
1222         .ndo_change_mtu =               netvsc_change_mtu,
1223         .ndo_validate_addr =            eth_validate_addr,
1224         .ndo_set_mac_address =          netvsc_set_mac_addr,
1225         .ndo_select_queue =             netvsc_select_queue,
1226         .ndo_get_stats64 =              netvsc_get_stats64,
1227 #ifdef CONFIG_NET_POLL_CONTROLLER
1228         .ndo_poll_controller =          netvsc_poll_controller,
1229 #endif
1230 };
1231
1232 /*
1233  * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
1234  * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
1235  * present send GARP packet to network peers with netif_notify_peers().
1236  */
1237 static void netvsc_link_change(struct work_struct *w)
1238 {
1239         struct net_device_context *ndev_ctx =
1240                 container_of(w, struct net_device_context, dwork.work);
1241         struct hv_device *device_obj = ndev_ctx->device_ctx;
1242         struct net_device *net = hv_get_drvdata(device_obj);
1243         struct netvsc_device *net_device;
1244         struct rndis_device *rdev;
1245         struct netvsc_reconfig *event = NULL;
1246         bool notify = false, reschedule = false;
1247         unsigned long flags, next_reconfig, delay;
1248
1249         rtnl_lock();
1250         net_device = rtnl_dereference(ndev_ctx->nvdev);
1251         if (!net_device)
1252                 goto out_unlock;
1253
1254         rdev = net_device->extension;
1255
1256         next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
1257         if (time_is_after_jiffies(next_reconfig)) {
1258                 /* link_watch only sends one notification with current state
1259                  * per second, avoid doing reconfig more frequently. Handle
1260                  * wrap around.
1261                  */
1262                 delay = next_reconfig - jiffies;
1263                 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
1264                 schedule_delayed_work(&ndev_ctx->dwork, delay);
1265                 goto out_unlock;
1266         }
1267         ndev_ctx->last_reconfig = jiffies;
1268
1269         spin_lock_irqsave(&ndev_ctx->lock, flags);
1270         if (!list_empty(&ndev_ctx->reconfig_events)) {
1271                 event = list_first_entry(&ndev_ctx->reconfig_events,
1272                                          struct netvsc_reconfig, list);
1273                 list_del(&event->list);
1274                 reschedule = !list_empty(&ndev_ctx->reconfig_events);
1275         }
1276         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1277
1278         if (!event)
1279                 goto out_unlock;
1280
1281         switch (event->event) {
1282                 /* Only the following events are possible due to the check in
1283                  * netvsc_linkstatus_callback()
1284                  */
1285         case RNDIS_STATUS_MEDIA_CONNECT:
1286                 if (rdev->link_state) {
1287                         rdev->link_state = false;
1288                         if (!ndev_ctx->datapath)
1289                                 netif_carrier_on(net);
1290                         netif_tx_wake_all_queues(net);
1291                 } else {
1292                         notify = true;
1293                 }
1294                 kfree(event);
1295                 break;
1296         case RNDIS_STATUS_MEDIA_DISCONNECT:
1297                 if (!rdev->link_state) {
1298                         rdev->link_state = true;
1299                         netif_carrier_off(net);
1300                         netif_tx_stop_all_queues(net);
1301                 }
1302                 kfree(event);
1303                 break;
1304         case RNDIS_STATUS_NETWORK_CHANGE:
1305                 /* Only makes sense if carrier is present */
1306                 if (!rdev->link_state) {
1307                         rdev->link_state = true;
1308                         netif_carrier_off(net);
1309                         netif_tx_stop_all_queues(net);
1310                         event->event = RNDIS_STATUS_MEDIA_CONNECT;
1311                         spin_lock_irqsave(&ndev_ctx->lock, flags);
1312                         list_add(&event->list, &ndev_ctx->reconfig_events);
1313                         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1314                         reschedule = true;
1315                 }
1316                 break;
1317         }
1318
1319         rtnl_unlock();
1320
1321         if (notify)
1322                 netdev_notify_peers(net);
1323
1324         /* link_watch only sends one notification with current state per
1325          * second, handle next reconfig event in 2 seconds.
1326          */
1327         if (reschedule)
1328                 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1329
1330         return;
1331
1332 out_unlock:
1333         rtnl_unlock();
1334 }
1335
1336 static struct net_device *get_netvsc_bymac(const u8 *mac)
1337 {
1338         struct net_device *dev;
1339
1340         ASSERT_RTNL();
1341
1342         for_each_netdev(&init_net, dev) {
1343                 if (dev->netdev_ops != &device_ops)
1344                         continue;       /* not a netvsc device */
1345
1346                 if (ether_addr_equal(mac, dev->perm_addr))
1347                         return dev;
1348         }
1349
1350         return NULL;
1351 }
1352
1353 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
1354 {
1355         struct net_device *dev;
1356
1357         ASSERT_RTNL();
1358
1359         for_each_netdev(&init_net, dev) {
1360                 struct net_device_context *net_device_ctx;
1361
1362                 if (dev->netdev_ops != &device_ops)
1363                         continue;       /* not a netvsc device */
1364
1365                 net_device_ctx = netdev_priv(dev);
1366                 if (net_device_ctx->nvdev == NULL)
1367                         continue;       /* device is removed */
1368
1369                 if (rtnl_dereference(net_device_ctx->vf_netdev) == vf_netdev)
1370                         return dev;     /* a match */
1371         }
1372
1373         return NULL;
1374 }
1375
1376 static int netvsc_register_vf(struct net_device *vf_netdev)
1377 {
1378         struct net_device *ndev;
1379         struct net_device_context *net_device_ctx;
1380         struct netvsc_device *netvsc_dev;
1381
1382         if (vf_netdev->addr_len != ETH_ALEN)
1383                 return NOTIFY_DONE;
1384
1385         /*
1386          * We will use the MAC address to locate the synthetic interface to
1387          * associate with the VF interface. If we don't find a matching
1388          * synthetic interface, move on.
1389          */
1390         ndev = get_netvsc_bymac(vf_netdev->perm_addr);
1391         if (!ndev)
1392                 return NOTIFY_DONE;
1393
1394         net_device_ctx = netdev_priv(ndev);
1395         netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
1396         if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
1397                 return NOTIFY_DONE;
1398
1399         netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
1400         /*
1401          * Take a reference on the module.
1402          */
1403         try_module_get(THIS_MODULE);
1404
1405         dev_hold(vf_netdev);
1406         rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
1407         return NOTIFY_OK;
1408 }
1409
1410 static int netvsc_vf_up(struct net_device *vf_netdev)
1411 {
1412         struct net_device *ndev;
1413         struct netvsc_device *netvsc_dev;
1414         struct net_device_context *net_device_ctx;
1415
1416         ndev = get_netvsc_byref(vf_netdev);
1417         if (!ndev)
1418                 return NOTIFY_DONE;
1419
1420         net_device_ctx = netdev_priv(ndev);
1421         netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
1422
1423         netdev_info(ndev, "VF up: %s\n", vf_netdev->name);
1424
1425         /*
1426          * Open the device before switching data path.
1427          */
1428         rndis_filter_open(netvsc_dev);
1429
1430         /*
1431          * notify the host to switch the data path.
1432          */
1433         netvsc_switch_datapath(ndev, true);
1434         netdev_info(ndev, "Data path switched to VF: %s\n", vf_netdev->name);
1435
1436         netif_carrier_off(ndev);
1437
1438         /* Now notify peers through VF device. */
1439         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, vf_netdev);
1440
1441         return NOTIFY_OK;
1442 }
1443
1444 static int netvsc_vf_down(struct net_device *vf_netdev)
1445 {
1446         struct net_device *ndev;
1447         struct netvsc_device *netvsc_dev;
1448         struct net_device_context *net_device_ctx;
1449
1450         ndev = get_netvsc_byref(vf_netdev);
1451         if (!ndev)
1452                 return NOTIFY_DONE;
1453
1454         net_device_ctx = netdev_priv(ndev);
1455         netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
1456
1457         netdev_info(ndev, "VF down: %s\n", vf_netdev->name);
1458         netvsc_switch_datapath(ndev, false);
1459         netdev_info(ndev, "Data path switched from VF: %s\n", vf_netdev->name);
1460         rndis_filter_close(netvsc_dev);
1461         netif_carrier_on(ndev);
1462
1463         /* Now notify peers through netvsc device. */
1464         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, ndev);
1465
1466         return NOTIFY_OK;
1467 }
1468
1469 static int netvsc_unregister_vf(struct net_device *vf_netdev)
1470 {
1471         struct net_device *ndev;
1472         struct net_device_context *net_device_ctx;
1473
1474         ndev = get_netvsc_byref(vf_netdev);
1475         if (!ndev)
1476                 return NOTIFY_DONE;
1477
1478         net_device_ctx = netdev_priv(ndev);
1479
1480         netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
1481
1482         RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
1483         dev_put(vf_netdev);
1484         module_put(THIS_MODULE);
1485         return NOTIFY_OK;
1486 }
1487
1488 static int netvsc_probe(struct hv_device *dev,
1489                         const struct hv_vmbus_device_id *dev_id)
1490 {
1491         struct net_device *net = NULL;
1492         struct net_device_context *net_device_ctx;
1493         struct netvsc_device_info device_info;
1494         struct netvsc_device *nvdev;
1495         int ret;
1496
1497         net = alloc_etherdev_mq(sizeof(struct net_device_context),
1498                                 VRSS_CHANNEL_MAX);
1499         if (!net)
1500                 return -ENOMEM;
1501
1502         netif_carrier_off(net);
1503
1504         netvsc_init_settings(net);
1505
1506         net_device_ctx = netdev_priv(net);
1507         net_device_ctx->device_ctx = dev;
1508         net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
1509         if (netif_msg_probe(net_device_ctx))
1510                 netdev_dbg(net, "netvsc msg_enable: %d\n",
1511                            net_device_ctx->msg_enable);
1512
1513         hv_set_drvdata(dev, net);
1514
1515         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
1516
1517         spin_lock_init(&net_device_ctx->lock);
1518         INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
1519
1520         net->netdev_ops = &device_ops;
1521         net->ethtool_ops = &ethtool_ops;
1522         SET_NETDEV_DEV(net, &dev->device);
1523
1524         /* We always need headroom for rndis header */
1525         net->needed_headroom = RNDIS_AND_PPI_SIZE;
1526
1527         /* Notify the netvsc driver of the new device */
1528         memset(&device_info, 0, sizeof(device_info));
1529         device_info.ring_size = ring_size;
1530         device_info.num_chn = VRSS_CHANNEL_DEFAULT;
1531         ret = rndis_filter_device_add(dev, &device_info);
1532         if (ret != 0) {
1533                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
1534                 free_netdev(net);
1535                 hv_set_drvdata(dev, NULL);
1536                 return ret;
1537         }
1538         memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
1539
1540         /* hw_features computed in rndis_filter_device_add */
1541         net->features = net->hw_features |
1542                 NETIF_F_HIGHDMA | NETIF_F_SG |
1543                 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
1544         net->vlan_features = net->features;
1545
1546         /* RCU not necessary here, device not registered */
1547         nvdev = net_device_ctx->nvdev;
1548         netif_set_real_num_tx_queues(net, nvdev->num_chn);
1549         netif_set_real_num_rx_queues(net, nvdev->num_chn);
1550
1551         /* MTU range: 68 - 1500 or 65521 */
1552         net->min_mtu = NETVSC_MTU_MIN;
1553         if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
1554                 net->max_mtu = NETVSC_MTU - ETH_HLEN;
1555         else
1556                 net->max_mtu = ETH_DATA_LEN;
1557
1558         ret = register_netdev(net);
1559         if (ret != 0) {
1560                 pr_err("Unable to register netdev.\n");
1561                 rndis_filter_device_remove(dev, nvdev);
1562                 free_netdev(net);
1563         }
1564
1565         return ret;
1566 }
1567
1568 static int netvsc_remove(struct hv_device *dev)
1569 {
1570         struct net_device *net;
1571         struct net_device_context *ndev_ctx;
1572
1573         net = hv_get_drvdata(dev);
1574
1575         if (net == NULL) {
1576                 dev_err(&dev->device, "No net device to remove\n");
1577                 return 0;
1578         }
1579
1580         ndev_ctx = netdev_priv(net);
1581
1582         netif_device_detach(net);
1583
1584         cancel_delayed_work_sync(&ndev_ctx->dwork);
1585
1586         /*
1587          * Call to the vsc driver to let it know that the device is being
1588          * removed. Also blocks mtu and channel changes.
1589          */
1590         rtnl_lock();
1591         rndis_filter_device_remove(dev, ndev_ctx->nvdev);
1592         rtnl_unlock();
1593
1594         unregister_netdev(net);
1595
1596         hv_set_drvdata(dev, NULL);
1597
1598         free_netdev(net);
1599         return 0;
1600 }
1601
1602 static const struct hv_vmbus_device_id id_table[] = {
1603         /* Network guid */
1604         { HV_NIC_GUID, },
1605         { },
1606 };
1607
1608 MODULE_DEVICE_TABLE(vmbus, id_table);
1609
1610 /* The one and only one */
1611 static struct  hv_driver netvsc_drv = {
1612         .name = KBUILD_MODNAME,
1613         .id_table = id_table,
1614         .probe = netvsc_probe,
1615         .remove = netvsc_remove,
1616 };
1617
1618 /*
1619  * On Hyper-V, every VF interface is matched with a corresponding
1620  * synthetic interface. The synthetic interface is presented first
1621  * to the guest. When the corresponding VF instance is registered,
1622  * we will take care of switching the data path.
1623  */
1624 static int netvsc_netdev_event(struct notifier_block *this,
1625                                unsigned long event, void *ptr)
1626 {
1627         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
1628
1629         /* Skip our own events */
1630         if (event_dev->netdev_ops == &device_ops)
1631                 return NOTIFY_DONE;
1632
1633         /* Avoid non-Ethernet type devices */
1634         if (event_dev->type != ARPHRD_ETHER)
1635                 return NOTIFY_DONE;
1636
1637         /* Avoid Vlan dev with same MAC registering as VF */
1638         if (is_vlan_dev(event_dev))
1639                 return NOTIFY_DONE;
1640
1641         /* Avoid Bonding master dev with same MAC registering as VF */
1642         if ((event_dev->priv_flags & IFF_BONDING) &&
1643             (event_dev->flags & IFF_MASTER))
1644                 return NOTIFY_DONE;
1645
1646         switch (event) {
1647         case NETDEV_REGISTER:
1648                 return netvsc_register_vf(event_dev);
1649         case NETDEV_UNREGISTER:
1650                 return netvsc_unregister_vf(event_dev);
1651         case NETDEV_UP:
1652                 return netvsc_vf_up(event_dev);
1653         case NETDEV_DOWN:
1654                 return netvsc_vf_down(event_dev);
1655         default:
1656                 return NOTIFY_DONE;
1657         }
1658 }
1659
1660 static struct notifier_block netvsc_netdev_notifier = {
1661         .notifier_call = netvsc_netdev_event,
1662 };
1663
1664 static void __exit netvsc_drv_exit(void)
1665 {
1666         unregister_netdevice_notifier(&netvsc_netdev_notifier);
1667         vmbus_driver_unregister(&netvsc_drv);
1668 }
1669
1670 static int __init netvsc_drv_init(void)
1671 {
1672         int ret;
1673
1674         if (ring_size < RING_SIZE_MIN) {
1675                 ring_size = RING_SIZE_MIN;
1676                 pr_info("Increased ring_size to %d (min allowed)\n",
1677                         ring_size);
1678         }
1679         ret = vmbus_driver_register(&netvsc_drv);
1680
1681         if (ret)
1682                 return ret;
1683
1684         register_netdevice_notifier(&netvsc_netdev_notifier);
1685         return 0;
1686 }
1687
1688 MODULE_LICENSE("GPL");
1689 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
1690
1691 module_init(netvsc_drv_init);
1692 module_exit(netvsc_drv_exit);