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
  41 #include "hyperv_net.h"
  42
  43 #define RING_SIZE_MIN 64
  44 #define LINKCHANGE_INT (2 * HZ)
  45 #define NETVSC_HW_FEATURES      (NETIF_F_RXCSUM | \
  46                                  NETIF_F_SG | \
  47                                  NETIF_F_TSO | \
  48                                  NETIF_F_TSO6 | \
  49                                  NETIF_F_HW_CSUM)
  50 static int ring_size = 128;
  51 module_param(ring_size, int, S_IRUGO);
  52 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
  53
  54 static int max_num_vrss_chns = 8;
  55
  56 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
  57                                 NETIF_MSG_LINK | NETIF_MSG_IFUP |
  58                                 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
  59                                 NETIF_MSG_TX_ERR;
  60
  61 static int debug = -1;
  62 module_param(debug, int, S_IRUGO);
  63 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  64
  65 static void do_set_multicast(struct work_struct *w)
  66 {
  67         struct net_device_context *ndevctx =
  68                 container_of(w, struct net_device_context, work);
  69         struct hv_device *device_obj = ndevctx->device_ctx;
  70         struct net_device *ndev = hv_get_drvdata(device_obj);
  71         struct netvsc_device *nvdev = ndevctx->nvdev;
  72         struct rndis_device *rdev;
  73
  74         if (!nvdev)
  75                 return;
  76
  77         rdev = nvdev->extension;
  78         if (rdev == NULL)
  79                 return;
  80
  81         if (ndev->flags & IFF_PROMISC)
  82                 rndis_filter_set_packet_filter(rdev,
  83                         NDIS_PACKET_TYPE_PROMISCUOUS);
  84         else
  85                 rndis_filter_set_packet_filter(rdev,
  86                         NDIS_PACKET_TYPE_BROADCAST |
  87                         NDIS_PACKET_TYPE_ALL_MULTICAST |
  88                         NDIS_PACKET_TYPE_DIRECTED);
  89 }
  90
  91 static void netvsc_set_multicast_list(struct net_device *net)
  92 {
  93         struct net_device_context *net_device_ctx = netdev_priv(net);
  94
  95         schedule_work(&net_device_ctx->work);
  96 }
  97
  98 static int netvsc_open(struct net_device *net)
  99 {
 100         struct netvsc_device *nvdev = net_device_to_netvsc_device(net);
 101         struct rndis_device *rdev;
 102         int ret = 0;
 103
 104         netif_carrier_off(net);
 105
 106         /* Open up the device */
 107         ret = rndis_filter_open(nvdev);
 108         if (ret != 0) {
 109                 netdev_err(net, "unable to open device (ret %d).\n", ret);
 110                 return ret;
 111         }
 112
 113         netif_tx_wake_all_queues(net);
 114
 115         rdev = nvdev->extension;
 116         if (!rdev->link_state)
 117                 netif_carrier_on(net);
 118
 119         return ret;
 120 }
 121
 122 static int netvsc_close(struct net_device *net)
 123 {
 124         struct net_device_context *net_device_ctx = netdev_priv(net);
 125         struct netvsc_device *nvdev = net_device_ctx->nvdev;
 126         int ret;
 127         u32 aread, awrite, i, msec = 10, retry = 0, retry_max = 20;
 128         struct vmbus_channel *chn;
 129
 130         netif_tx_disable(net);
 131
 132         /* Make sure netvsc_set_multicast_list doesn't re-enable filter! */
 133         cancel_work_sync(&net_device_ctx->work);
 134         ret = rndis_filter_close(nvdev);
 135         if (ret != 0) {
 136                 netdev_err(net, "unable to close device (ret %d).\n", ret);
 137                 return ret;
 138         }
 139
 140         /* Ensure pending bytes in ring are read */
 141         while (true) {
 142                 aread = 0;
 143                 for (i = 0; i < nvdev->num_chn; i++) {
 144                         chn = nvdev->chn_table[i];
 145                         if (!chn)
 146                                 continue;
 147
 148                         hv_get_ringbuffer_availbytes(&chn->inbound, &aread,
 149                                                      &awrite);
 150
 151                         if (aread)
 152                                 break;
 153
 154                         hv_get_ringbuffer_availbytes(&chn->outbound, &aread,
 155                                                      &awrite);
 156
 157                         if (aread)
 158                                 break;
 159                 }
 160
 161                 retry++;
 162                 if (retry > retry_max || aread == 0)
 163                         break;
 164
 165                 msleep(msec);
 166
 167                 if (msec < 1000)
 168                         msec *= 2;
 169         }
 170
 171         if (aread) {
 172                 netdev_err(net, "Ring buffer not empty after closing rndis\n");
 173                 ret = -ETIMEDOUT;
 174         }
 175
 176         return ret;
 177 }
 178
 179 static void *init_ppi_data(struct rndis_message *msg, u32 ppi_size,
 180                                 int pkt_type)
 181 {
 182         struct rndis_packet *rndis_pkt;
 183         struct rndis_per_packet_info *ppi;
 184
 185         rndis_pkt = &msg->msg.pkt;
 186         rndis_pkt->data_offset += ppi_size;
 187
 188         ppi = (struct rndis_per_packet_info *)((void *)rndis_pkt +
 189                 rndis_pkt->per_pkt_info_offset + rndis_pkt->per_pkt_info_len);
 190
 191         ppi->size = ppi_size;
 192         ppi->type = pkt_type;
 193         ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
 194
 195         rndis_pkt->per_pkt_info_len += ppi_size;
 196
 197         return ppi;
 198 }
 199
 200 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
 201                         void *accel_priv, select_queue_fallback_t fallback)
 202 {
 203         struct net_device_context *net_device_ctx = netdev_priv(ndev);
 204         struct netvsc_device *nvsc_dev = net_device_ctx->nvdev;
 205         u32 hash;
 206         u16 q_idx = 0;
 207
 208         if (nvsc_dev == NULL || ndev->real_num_tx_queues <= 1)
 209                 return 0;
 210
 211         hash = skb_get_hash(skb);
 212         q_idx = nvsc_dev->send_table[hash % VRSS_SEND_TAB_SIZE] %
 213                 ndev->real_num_tx_queues;
 214
 215         if (!nvsc_dev->chn_table[q_idx])
 216                 q_idx = 0;
 217
 218         return q_idx;
 219 }
 220
 221 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
 222                         struct hv_page_buffer *pb)
 223 {
 224         int j = 0;
 225
 226         /* Deal with compund pages by ignoring unused part
 227          * of the page.
 228          */
 229         page += (offset >> PAGE_SHIFT);
 230         offset &= ~PAGE_MASK;
 231
 232         while (len > 0) {
 233                 unsigned long bytes;
 234
 235                 bytes = PAGE_SIZE - offset;
 236                 if (bytes > len)
 237                         bytes = len;
 238                 pb[j].pfn = page_to_pfn(page);
 239                 pb[j].offset = offset;
 240                 pb[j].len = bytes;
 241
 242                 offset += bytes;
 243                 len -= bytes;
 244
 245                 if (offset == PAGE_SIZE && len) {
 246                         page++;
 247                         offset = 0;
 248                         j++;
 249                 }
 250         }
 251
 252         return j + 1;
 253 }
 254
 255 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
 256                            struct hv_netvsc_packet *packet,
 257                            struct hv_page_buffer **page_buf)
 258 {
 259         struct hv_page_buffer *pb = *page_buf;
 260         u32 slots_used = 0;
 261         char *data = skb->data;
 262         int frags = skb_shinfo(skb)->nr_frags;
 263         int i;
 264
 265         /* The packet is laid out thus:
 266          * 1. hdr: RNDIS header and PPI
 267          * 2. skb linear data
 268          * 3. skb fragment data
 269          */
 270         if (hdr != NULL)
 271                 slots_used += fill_pg_buf(virt_to_page(hdr),
 272                                         offset_in_page(hdr),
 273                                         len, &pb[slots_used]);
 274
 275         packet->rmsg_size = len;
 276         packet->rmsg_pgcnt = slots_used;
 277
 278         slots_used += fill_pg_buf(virt_to_page(data),
 279                                 offset_in_page(data),
 280                                 skb_headlen(skb), &pb[slots_used]);
 281
 282         for (i = 0; i < frags; i++) {
 283                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 284
 285                 slots_used += fill_pg_buf(skb_frag_page(frag),
 286                                         frag->page_offset,
 287                                         skb_frag_size(frag), &pb[slots_used]);
 288         }
 289         return slots_used;
 290 }
 291
 292 static int count_skb_frag_slots(struct sk_buff *skb)
 293 {
 294         int i, frags = skb_shinfo(skb)->nr_frags;
 295         int pages = 0;
 296
 297         for (i = 0; i < frags; i++) {
 298                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 299                 unsigned long size = skb_frag_size(frag);
 300                 unsigned long offset = frag->page_offset;
 301
 302                 /* Skip unused frames from start of page */
 303                 offset &= ~PAGE_MASK;
 304                 pages += PFN_UP(offset + size);
 305         }
 306         return pages;
 307 }
 308
 309 static int netvsc_get_slots(struct sk_buff *skb)
 310 {
 311         char *data = skb->data;
 312         unsigned int offset = offset_in_page(data);
 313         unsigned int len = skb_headlen(skb);
 314         int slots;
 315         int frag_slots;
 316
 317         slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
 318         frag_slots = count_skb_frag_slots(skb);
 319         return slots + frag_slots;
 320 }
 321
 322 static u32 get_net_transport_info(struct sk_buff *skb, u32 *trans_off)
 323 {
 324         u32 ret_val = TRANSPORT_INFO_NOT_IP;
 325
 326         if ((eth_hdr(skb)->h_proto != htons(ETH_P_IP)) &&
 327                 (eth_hdr(skb)->h_proto != htons(ETH_P_IPV6))) {
 328                 goto not_ip;
 329         }
 330
 331         *trans_off = skb_transport_offset(skb);
 332
 333         if ((eth_hdr(skb)->h_proto == htons(ETH_P_IP))) {
 334                 struct iphdr *iphdr = ip_hdr(skb);
 335
 336                 if (iphdr->protocol == IPPROTO_TCP)
 337                         ret_val = TRANSPORT_INFO_IPV4_TCP;
 338                 else if (iphdr->protocol == IPPROTO_UDP)
 339                         ret_val = TRANSPORT_INFO_IPV4_UDP;
 340         } else {
 341                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
 342                         ret_val = TRANSPORT_INFO_IPV6_TCP;
 343                 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
 344                         ret_val = TRANSPORT_INFO_IPV6_UDP;
 345         }
 346
 347 not_ip:
 348         return ret_val;
 349 }
 350
 351 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
 352 {
 353         struct net_device_context *net_device_ctx = netdev_priv(net);
 354         struct hv_netvsc_packet *packet = NULL;
 355         int ret;
 356         unsigned int num_data_pgs;
 357         struct rndis_message *rndis_msg;
 358         struct rndis_packet *rndis_pkt;
 359         u32 rndis_msg_size;
 360         struct rndis_per_packet_info *ppi;
 361         struct ndis_tcp_ip_checksum_info *csum_info;
 362         int  hdr_offset;
 363         u32 net_trans_info;
 364         u32 hash;
 365         u32 skb_length;
 366         struct hv_page_buffer page_buf[MAX_PAGE_BUFFER_COUNT];
 367         struct hv_page_buffer *pb = page_buf;
 368
 369         /* We will atmost need two pages to describe the rndis
 370          * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
 371          * of pages in a single packet. If skb is scattered around
 372          * more pages we try linearizing it.
 373          */
 374
 375         skb_length = skb->len;
 376         num_data_pgs = netvsc_get_slots(skb) + 2;
 377
 378         if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
 379                 ++net_device_ctx->eth_stats.tx_scattered;
 380
 381                 if (skb_linearize(skb))
 382                         goto no_memory;
 383
 384                 num_data_pgs = netvsc_get_slots(skb) + 2;
 385                 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
 386                         ++net_device_ctx->eth_stats.tx_too_big;
 387                         goto drop;
 388                 }
 389         }
 390
 391         /*
 392          * Place the rndis header in the skb head room and
 393          * the skb->cb will be used for hv_netvsc_packet
 394          * structure.
 395          */
 396         ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
 397         if (ret)
 398                 goto no_memory;
 399
 400         /* Use the skb control buffer for building up the packet */
 401         BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
 402                         FIELD_SIZEOF(struct sk_buff, cb));
 403         packet = (struct hv_netvsc_packet *)skb->cb;
 404
 405         packet->q_idx = skb_get_queue_mapping(skb);
 406
 407         packet->total_data_buflen = skb->len;
 408
 409         rndis_msg = (struct rndis_message *)skb->head;
 410
 411         memset(rndis_msg, 0, RNDIS_AND_PPI_SIZE);
 412
 413         /* Add the rndis header */
 414         rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
 415         rndis_msg->msg_len = packet->total_data_buflen;
 416         rndis_pkt = &rndis_msg->msg.pkt;
 417         rndis_pkt->data_offset = sizeof(struct rndis_packet);
 418         rndis_pkt->data_len = packet->total_data_buflen;
 419         rndis_pkt->per_pkt_info_offset = sizeof(struct rndis_packet);
 420
 421         rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
 422
 423         hash = skb_get_hash_raw(skb);
 424         if (hash != 0 && net->real_num_tx_queues > 1) {
 425                 rndis_msg_size += NDIS_HASH_PPI_SIZE;
 426                 ppi = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
 427                                     NBL_HASH_VALUE);
 428                 *(u32 *)((void *)ppi + ppi->ppi_offset) = hash;
 429         }
 430
 431         if (skb_vlan_tag_present(skb)) {
 432                 struct ndis_pkt_8021q_info *vlan;
 433
 434                 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
 435                 ppi = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
 436                                         IEEE_8021Q_INFO);
 437                 vlan = (struct ndis_pkt_8021q_info *)((void *)ppi +
 438                                                 ppi->ppi_offset);
 439                 vlan->vlanid = skb->vlan_tci & VLAN_VID_MASK;
 440                 vlan->pri = (skb->vlan_tci & VLAN_PRIO_MASK) >>
 441                                 VLAN_PRIO_SHIFT;
 442         }
 443
 444         net_trans_info = get_net_transport_info(skb, &hdr_offset);
 445
 446         /*
 447          * Setup the sendside checksum offload only if this is not a
 448          * GSO packet.
 449          */
 450         if ((net_trans_info & (INFO_TCP | INFO_UDP)) && skb_is_gso(skb)) {
 451                 struct ndis_tcp_lso_info *lso_info;
 452
 453                 rndis_msg_size += NDIS_LSO_PPI_SIZE;
 454                 ppi = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
 455                                     TCP_LARGESEND_PKTINFO);
 456
 457                 lso_info = (struct ndis_tcp_lso_info *)((void *)ppi +
 458                                                         ppi->ppi_offset);
 459
 460                 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
 461                 if (net_trans_info & (INFO_IPV4 << 16)) {
 462                         lso_info->lso_v2_transmit.ip_version =
 463                                 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
 464                         ip_hdr(skb)->tot_len = 0;
 465                         ip_hdr(skb)->check = 0;
 466                         tcp_hdr(skb)->check =
 467                                 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
 468                                                    ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 469                 } else {
 470                         lso_info->lso_v2_transmit.ip_version =
 471                                 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
 472                         ipv6_hdr(skb)->payload_len = 0;
 473                         tcp_hdr(skb)->check =
 474                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 475                                                  &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 476                 }
 477                 lso_info->lso_v2_transmit.tcp_header_offset = hdr_offset;
 478                 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
 479         } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
 480                 if (net_trans_info & INFO_TCP) {
 481                         rndis_msg_size += NDIS_CSUM_PPI_SIZE;
 482                         ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
 483                                             TCPIP_CHKSUM_PKTINFO);
 484
 485                         csum_info = (struct ndis_tcp_ip_checksum_info *)((void *)ppi +
 486                                                                          ppi->ppi_offset);
 487
 488                         if (net_trans_info & (INFO_IPV4 << 16))
 489                                 csum_info->transmit.is_ipv4 = 1;
 490                         else
 491                                 csum_info->transmit.is_ipv6 = 1;
 492
 493                         csum_info->transmit.tcp_checksum = 1;
 494                         csum_info->transmit.tcp_header_offset = hdr_offset;
 495                 } else {
 496                         /* UDP checksum (and other) offload is not supported. */
 497                         if (skb_checksum_help(skb))
 498                                 goto drop;
 499                 }
 500         }
 501
 502         /* Start filling in the page buffers with the rndis hdr */
 503         rndis_msg->msg_len += rndis_msg_size;
 504         packet->total_data_buflen = rndis_msg->msg_len;
 505         packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
 506                                                skb, packet, &pb);
 507
 508         /* timestamp packet in software */
 509         skb_tx_timestamp(skb);
 510         ret = netvsc_send(net_device_ctx->device_ctx, packet,
 511                           rndis_msg, &pb, skb);
 512         if (likely(ret == 0)) {
 513                 struct netvsc_stats *tx_stats = this_cpu_ptr(net_device_ctx->tx_stats);
 514
 515                 u64_stats_update_begin(&tx_stats->syncp);
 516                 tx_stats->packets++;
 517                 tx_stats->bytes += skb_length;
 518                 u64_stats_update_end(&tx_stats->syncp);
 519                 return NETDEV_TX_OK;
 520         }
 521
 522         if (ret == -EAGAIN) {
 523                 ++net_device_ctx->eth_stats.tx_busy;
 524                 return NETDEV_TX_BUSY;
 525         }
 526
 527         if (ret == -ENOSPC)
 528                 ++net_device_ctx->eth_stats.tx_no_space;
 529
 530 drop:
 531         dev_kfree_skb_any(skb);
 532         net->stats.tx_dropped++;
 533
 534         return NETDEV_TX_OK;
 535
 536 no_memory:
 537         ++net_device_ctx->eth_stats.tx_no_memory;
 538         goto drop;
 539 }
 540
 541 /*
 542  * netvsc_linkstatus_callback - Link up/down notification
 543  */
 544 void netvsc_linkstatus_callback(struct hv_device *device_obj,
 545                                 struct rndis_message *resp)
 546 {
 547         struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
 548         struct net_device *net;
 549         struct net_device_context *ndev_ctx;
 550         struct netvsc_reconfig *event;
 551         unsigned long flags;
 552
 553         net = hv_get_drvdata(device_obj);
 554
 555         if (!net)
 556                 return;
 557
 558         ndev_ctx = netdev_priv(net);
 559
 560         /* Update the physical link speed when changing to another vSwitch */
 561         if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
 562                 u32 speed;
 563
 564                 speed = *(u32 *)((void *)indicate + indicate->
 565                                  status_buf_offset) / 10000;
 566                 ndev_ctx->speed = speed;
 567                 return;
 568         }
 569
 570         /* Handle these link change statuses below */
 571         if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
 572             indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
 573             indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
 574                 return;
 575
 576         if (net->reg_state != NETREG_REGISTERED)
 577                 return;
 578
 579         event = kzalloc(sizeof(*event), GFP_ATOMIC);
 580         if (!event)
 581                 return;
 582         event->event = indicate->status;
 583
 584         spin_lock_irqsave(&ndev_ctx->lock, flags);
 585         list_add_tail(&event->list, &ndev_ctx->reconfig_events);
 586         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
 587
 588         schedule_delayed_work(&ndev_ctx->dwork, 0);
 589 }
 590
 591 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
 592                                 struct hv_netvsc_packet *packet,
 593                                 struct ndis_tcp_ip_checksum_info *csum_info,
 594                                 void *data, u16 vlan_tci)
 595 {
 596         struct sk_buff *skb;
 597
 598         skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
 599         if (!skb)
 600                 return skb;
 601
 602         /*
 603          * Copy to skb. This copy is needed here since the memory pointed by
 604          * hv_netvsc_packet cannot be deallocated
 605          */
 606         memcpy(skb_put(skb, packet->total_data_buflen), data,
 607                packet->total_data_buflen);
 608
 609         skb->protocol = eth_type_trans(skb, net);
 610
 611         /* skb is already created with CHECKSUM_NONE */
 612         skb_checksum_none_assert(skb);
 613
 614         /*
 615          * In Linux, the IP checksum is always checked.
 616          * Do L4 checksum offload if enabled and present.
 617          */
 618         if (csum_info && (net->features & NETIF_F_RXCSUM)) {
 619                 if (csum_info->receive.tcp_checksum_succeeded ||
 620                     csum_info->receive.udp_checksum_succeeded)
 621                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 622         }
 623
 624         if (vlan_tci & VLAN_TAG_PRESENT)
 625                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 626                                        vlan_tci);
 627
 628         return skb;
 629 }
 630
 631 /*
 632  * netvsc_recv_callback -  Callback when we receive a packet from the
 633  * "wire" on the specified device.
 634  */
 635 int netvsc_recv_callback(struct hv_device *device_obj,
 636                                 struct hv_netvsc_packet *packet,
 637                                 void **data,
 638                                 struct ndis_tcp_ip_checksum_info *csum_info,
 639                                 struct vmbus_channel *channel,
 640                                 u16 vlan_tci)
 641 {
 642         struct net_device *net = hv_get_drvdata(device_obj);
 643         struct net_device_context *net_device_ctx = netdev_priv(net);
 644         struct net_device *vf_netdev;
 645         struct sk_buff *skb;
 646         struct netvsc_stats *rx_stats;
 647
 648         if (net->reg_state != NETREG_REGISTERED)
 649                 return NVSP_STAT_FAIL;
 650
 651         /*
 652          * If necessary, inject this packet into the VF interface.
 653          * On Hyper-V, multicast and brodcast packets are only delivered
 654          * to the synthetic interface (after subjecting these to
 655          * policy filters on the host). Deliver these via the VF
 656          * interface in the guest.
 657          */
 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, packet, csum_info, *data, vlan_tci);
 664         if (unlikely(!skb)) {
 665                 ++net->stats.rx_dropped;
 666                 return NVSP_STAT_FAIL;
 667         }
 668
 669         if (net != vf_netdev)
 670                 skb_record_rx_queue(skb,
 671                                     channel->offermsg.offer.sub_channel_index);
 672
 673         /*
 674          * Even if injecting the packet, record the statistics
 675          * on the synthetic device because modifying the VF device
 676          * statistics will not work correctly.
 677          */
 678         rx_stats = this_cpu_ptr(net_device_ctx->rx_stats);
 679         u64_stats_update_begin(&rx_stats->syncp);
 680         rx_stats->packets++;
 681         rx_stats->bytes += packet->total_data_buflen;
 682
 683         if (skb->pkt_type == PACKET_BROADCAST)
 684                 ++rx_stats->broadcast;
 685         else if (skb->pkt_type == PACKET_MULTICAST)
 686                 ++rx_stats->multicast;
 687         u64_stats_update_end(&rx_stats->syncp);
 688
 689         /*
 690          * Pass the skb back up. Network stack will deallocate the skb when it
 691          * is done.
 692          * TODO - use NAPI?
 693          */
 694         netif_rx(skb);
 695
 696         return 0;
 697 }
 698
 699 static void netvsc_get_drvinfo(struct net_device *net,
 700                                struct ethtool_drvinfo *info)
 701 {
 702         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
 703         strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
 704 }
 705
 706 static void netvsc_get_channels(struct net_device *net,
 707                                 struct ethtool_channels *channel)
 708 {
 709         struct net_device_context *net_device_ctx = netdev_priv(net);
 710         struct netvsc_device *nvdev = net_device_ctx->nvdev;
 711
 712         if (nvdev) {
 713                 channel->max_combined   = nvdev->max_chn;
 714                 channel->combined_count = nvdev->num_chn;
 715         }
 716 }
 717
 718 static int netvsc_set_channels(struct net_device *net,
 719                                struct ethtool_channels *channels)
 720 {
 721         struct net_device_context *net_device_ctx = netdev_priv(net);
 722         struct hv_device *dev = net_device_ctx->device_ctx;
 723         struct netvsc_device *nvdev = net_device_ctx->nvdev;
 724         struct netvsc_device_info device_info;
 725         u32 num_chn;
 726         u32 max_chn;
 727         int ret = 0;
 728         bool recovering = false;
 729
 730         if (net_device_ctx->start_remove || !nvdev || nvdev->destroy)
 731                 return -ENODEV;
 732
 733         num_chn = nvdev->num_chn;
 734         max_chn = min_t(u32, nvdev->max_chn, num_online_cpus());
 735
 736         if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5) {
 737                 pr_info("vRSS unsupported before NVSP Version 5\n");
 738                 return -EINVAL;
 739         }
 740
 741         /* We do not support rx, tx, or other */
 742         if (!channels ||
 743             channels->rx_count ||
 744             channels->tx_count ||
 745             channels->other_count ||
 746             (channels->combined_count < 1))
 747                 return -EINVAL;
 748
 749         if (channels->combined_count > max_chn) {
 750                 pr_info("combined channels too high, using %d\n", max_chn);
 751                 channels->combined_count = max_chn;
 752         }
 753
 754         ret = netvsc_close(net);
 755         if (ret)
 756                 goto out;
 757
 758  do_set:
 759         net_device_ctx->start_remove = true;
 760         rndis_filter_device_remove(dev);
 761
 762         nvdev->num_chn = channels->combined_count;
 763
 764         memset(&device_info, 0, sizeof(device_info));
 765         device_info.num_chn = nvdev->num_chn; /* passed to RNDIS */
 766         device_info.ring_size = ring_size;
 767         device_info.max_num_vrss_chns = max_num_vrss_chns;
 768
 769         ret = rndis_filter_device_add(dev, &device_info);
 770         if (ret) {
 771                 if (recovering) {
 772                         netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
 773                         return ret;
 774                 }
 775                 goto recover;
 776         }
 777
 778         nvdev = net_device_ctx->nvdev;
 779
 780         ret = netif_set_real_num_tx_queues(net, nvdev->num_chn);
 781         if (ret) {
 782                 if (recovering) {
 783                         netdev_err(net, "could not set tx queue count (ret %d)\n", ret);
 784                         return ret;
 785                 }
 786                 goto recover;
 787         }
 788
 789         ret = netif_set_real_num_rx_queues(net, nvdev->num_chn);
 790         if (ret) {
 791                 if (recovering) {
 792                         netdev_err(net, "could not set rx queue count (ret %d)\n", ret);
 793                         return ret;
 794                 }
 795                 goto recover;
 796         }
 797
 798  out:
 799         netvsc_open(net);
 800         net_device_ctx->start_remove = false;
 801         /* We may have missed link change notifications */
 802         schedule_delayed_work(&net_device_ctx->dwork, 0);
 803
 804         return ret;
 805
 806  recover:
 807         /* If the above failed, we attempt to recover through the same
 808          * process but with the original number of channels.
 809          */
 810         netdev_err(net, "could not set channels, recovering\n");
 811         recovering = true;
 812         channels->combined_count = num_chn;
 813         goto do_set;
 814 }
 815
 816 static bool netvsc_validate_ethtool_ss_cmd(const struct ethtool_cmd *cmd)
 817 {
 818         struct ethtool_cmd diff1 = *cmd;
 819         struct ethtool_cmd diff2 = {};
 820
 821         ethtool_cmd_speed_set(&diff1, 0);
 822         diff1.duplex = 0;
 823         /* advertising and cmd are usually set */
 824         diff1.advertising = 0;
 825         diff1.cmd = 0;
 826         /* We set port to PORT_OTHER */
 827         diff2.port = PORT_OTHER;
 828
 829         return !memcmp(&diff1, &diff2, sizeof(diff1));
 830 }
 831
 832 static void netvsc_init_settings(struct net_device *dev)
 833 {
 834         struct net_device_context *ndc = netdev_priv(dev);
 835
 836         ndc->speed = SPEED_UNKNOWN;
 837         ndc->duplex = DUPLEX_UNKNOWN;
 838 }
 839
 840 static int netvsc_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
 841 {
 842         struct net_device_context *ndc = netdev_priv(dev);
 843
 844         ethtool_cmd_speed_set(cmd, ndc->speed);
 845         cmd->duplex = ndc->duplex;
 846         cmd->port = PORT_OTHER;
 847
 848         return 0;
 849 }
 850
 851 static int netvsc_set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
 852 {
 853         struct net_device_context *ndc = netdev_priv(dev);
 854         u32 speed;
 855
 856         speed = ethtool_cmd_speed(cmd);
 857         if (!ethtool_validate_speed(speed) ||
 858             !ethtool_validate_duplex(cmd->duplex) ||
 859             !netvsc_validate_ethtool_ss_cmd(cmd))
 860                 return -EINVAL;
 861
 862         ndc->speed = speed;
 863         ndc->duplex = cmd->duplex;
 864
 865         return 0;
 866 }
 867
 868 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
 869 {
 870         struct net_device_context *ndevctx = netdev_priv(ndev);
 871         struct netvsc_device *nvdev = ndevctx->nvdev;
 872         struct hv_device *hdev = ndevctx->device_ctx;
 873         struct netvsc_device_info device_info;
 874         int limit = ETH_DATA_LEN;
 875         u32 num_chn;
 876         int ret = 0;
 877
 878         if (ndevctx->start_remove || !nvdev || nvdev->destroy)
 879                 return -ENODEV;
 880
 881         if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
 882                 limit = NETVSC_MTU - ETH_HLEN;
 883
 884         if (mtu < NETVSC_MTU_MIN || mtu > limit)
 885                 return -EINVAL;
 886
 887         ret = netvsc_close(ndev);
 888         if (ret)
 889                 goto out;
 890
 891         num_chn = nvdev->num_chn;
 892
 893         ndevctx->start_remove = true;
 894         rndis_filter_device_remove(hdev);
 895
 896         ndev->mtu = mtu;
 897
 898         memset(&device_info, 0, sizeof(device_info));
 899         device_info.ring_size = ring_size;
 900         device_info.num_chn = num_chn;
 901         device_info.max_num_vrss_chns = max_num_vrss_chns;
 902         rndis_filter_device_add(hdev, &device_info);
 903
 904 out:
 905         netvsc_open(ndev);
 906         ndevctx->start_remove = false;
 907
 908         /* We may have missed link change notifications */
 909         schedule_delayed_work(&ndevctx->dwork, 0);
 910
 911         return ret;
 912 }
 913
 914 static struct rtnl_link_stats64 *netvsc_get_stats64(struct net_device *net,
 915                                                     struct rtnl_link_stats64 *t)
 916 {
 917         struct net_device_context *ndev_ctx = netdev_priv(net);
 918         int cpu;
 919
 920         for_each_possible_cpu(cpu) {
 921                 struct netvsc_stats *tx_stats = per_cpu_ptr(ndev_ctx->tx_stats,
 922                                                             cpu);
 923                 struct netvsc_stats *rx_stats = per_cpu_ptr(ndev_ctx->rx_stats,
 924                                                             cpu);
 925                 u64 tx_packets, tx_bytes, rx_packets, rx_bytes, rx_multicast;
 926                 unsigned int start;
 927
 928                 do {
 929                         start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
 930                         tx_packets = tx_stats->packets;
 931                         tx_bytes = tx_stats->bytes;
 932                 } while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
 933
 934                 do {
 935                         start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
 936                         rx_packets = rx_stats->packets;
 937                         rx_bytes = rx_stats->bytes;
 938                         rx_multicast = rx_stats->multicast + rx_stats->broadcast;
 939                 } while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
 940
 941                 t->tx_bytes     += tx_bytes;
 942                 t->tx_packets   += tx_packets;
 943                 t->rx_bytes     += rx_bytes;
 944                 t->rx_packets   += rx_packets;
 945                 t->multicast    += rx_multicast;
 946         }
 947
 948         t->tx_dropped   = net->stats.tx_dropped;
 949         t->tx_errors    = net->stats.tx_dropped;
 950
 951         t->rx_dropped   = net->stats.rx_dropped;
 952         t->rx_errors    = net->stats.rx_errors;
 953
 954         return t;
 955 }
 956
 957 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
 958 {
 959         struct sockaddr *addr = p;
 960         char save_adr[ETH_ALEN];
 961         unsigned char save_aatype;
 962         int err;
 963
 964         memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
 965         save_aatype = ndev->addr_assign_type;
 966
 967         err = eth_mac_addr(ndev, p);
 968         if (err != 0)
 969                 return err;
 970
 971         err = rndis_filter_set_device_mac(ndev, addr->sa_data);
 972         if (err != 0) {
 973                 /* roll back to saved MAC */
 974                 memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
 975                 ndev->addr_assign_type = save_aatype;
 976         }
 977
 978         return err;
 979 }
 980
 981 static const struct {
 982         char name[ETH_GSTRING_LEN];
 983         u16 offset;
 984 } netvsc_stats[] = {
 985         { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
 986         { "tx_no_memory",  offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
 987         { "tx_no_space",  offsetof(struct netvsc_ethtool_stats, tx_no_space) },
 988         { "tx_too_big",   offsetof(struct netvsc_ethtool_stats, tx_too_big) },
 989         { "tx_busy",      offsetof(struct netvsc_ethtool_stats, tx_busy) },
 990 };
 991
 992 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
 993 {
 994         switch (string_set) {
 995         case ETH_SS_STATS:
 996                 return ARRAY_SIZE(netvsc_stats);
 997         default:
 998                 return -EINVAL;
 999         }
1000 }
1001
1002 static void netvsc_get_ethtool_stats(struct net_device *dev,
1003                                      struct ethtool_stats *stats, u64 *data)
1004 {
1005         struct net_device_context *ndc = netdev_priv(dev);
1006         const void *nds = &ndc->eth_stats;
1007         int i;
1008
1009         for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++)
1010                 data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1011 }
1012
1013 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1014 {
1015         int i;
1016
1017         switch (stringset) {
1018         case ETH_SS_STATS:
1019                 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++)
1020                         memcpy(data + i * ETH_GSTRING_LEN,
1021                                netvsc_stats[i].name, ETH_GSTRING_LEN);
1022                 break;
1023         }
1024 }
1025
1026 #ifdef CONFIG_NET_POLL_CONTROLLER
1027 static void netvsc_poll_controller(struct net_device *net)
1028 {
1029         /* As netvsc_start_xmit() works synchronous we don't have to
1030          * trigger anything here.
1031          */
1032 }
1033 #endif
1034
1035 static const struct ethtool_ops ethtool_ops = {
1036         .get_drvinfo    = netvsc_get_drvinfo,
1037         .get_link       = ethtool_op_get_link,
1038         .get_ethtool_stats = netvsc_get_ethtool_stats,
1039         .get_sset_count = netvsc_get_sset_count,
1040         .get_strings    = netvsc_get_strings,
1041         .get_channels   = netvsc_get_channels,
1042         .set_channels   = netvsc_set_channels,
1043         .get_ts_info    = ethtool_op_get_ts_info,
1044         .get_settings   = netvsc_get_settings,
1045         .set_settings   = netvsc_set_settings,
1046 };
1047
1048 static const struct net_device_ops device_ops = {
1049         .ndo_open =                     netvsc_open,
1050         .ndo_stop =                     netvsc_close,
1051         .ndo_start_xmit =               netvsc_start_xmit,
1052         .ndo_set_rx_mode =              netvsc_set_multicast_list,
1053         .ndo_change_mtu =               netvsc_change_mtu,
1054         .ndo_validate_addr =            eth_validate_addr,
1055         .ndo_set_mac_address =          netvsc_set_mac_addr,
1056         .ndo_select_queue =             netvsc_select_queue,
1057         .ndo_get_stats64 =              netvsc_get_stats64,
1058 #ifdef CONFIG_NET_POLL_CONTROLLER
1059         .ndo_poll_controller =          netvsc_poll_controller,
1060 #endif
1061 };
1062
1063 /*
1064  * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
1065  * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
1066  * present send GARP packet to network peers with netif_notify_peers().
1067  */
1068 static void netvsc_link_change(struct work_struct *w)
1069 {
1070         struct net_device_context *ndev_ctx =
1071                 container_of(w, struct net_device_context, dwork.work);
1072         struct hv_device *device_obj = ndev_ctx->device_ctx;
1073         struct net_device *net = hv_get_drvdata(device_obj);
1074         struct netvsc_device *net_device;
1075         struct rndis_device *rdev;
1076         struct netvsc_reconfig *event = NULL;
1077         bool notify = false, reschedule = false;
1078         unsigned long flags, next_reconfig, delay;
1079
1080         rtnl_lock();
1081         if (ndev_ctx->start_remove)
1082                 goto out_unlock;
1083
1084         net_device = ndev_ctx->nvdev;
1085         rdev = net_device->extension;
1086
1087         next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
1088         if (time_is_after_jiffies(next_reconfig)) {
1089                 /* link_watch only sends one notification with current state
1090                  * per second, avoid doing reconfig more frequently. Handle
1091                  * wrap around.
1092                  */
1093                 delay = next_reconfig - jiffies;
1094                 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
1095                 schedule_delayed_work(&ndev_ctx->dwork, delay);
1096                 goto out_unlock;
1097         }
1098         ndev_ctx->last_reconfig = jiffies;
1099
1100         spin_lock_irqsave(&ndev_ctx->lock, flags);
1101         if (!list_empty(&ndev_ctx->reconfig_events)) {
1102                 event = list_first_entry(&ndev_ctx->reconfig_events,
1103                                          struct netvsc_reconfig, list);
1104                 list_del(&event->list);
1105                 reschedule = !list_empty(&ndev_ctx->reconfig_events);
1106         }
1107         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1108
1109         if (!event)
1110                 goto out_unlock;
1111
1112         switch (event->event) {
1113                 /* Only the following events are possible due to the check in
1114                  * netvsc_linkstatus_callback()
1115                  */
1116         case RNDIS_STATUS_MEDIA_CONNECT:
1117                 if (rdev->link_state) {
1118                         rdev->link_state = false;
1119                         netif_carrier_on(net);
1120                         netif_tx_wake_all_queues(net);
1121                 } else {
1122                         notify = true;
1123                 }
1124                 kfree(event);
1125                 break;
1126         case RNDIS_STATUS_MEDIA_DISCONNECT:
1127                 if (!rdev->link_state) {
1128                         rdev->link_state = true;
1129                         netif_carrier_off(net);
1130                         netif_tx_stop_all_queues(net);
1131                 }
1132                 kfree(event);
1133                 break;
1134         case RNDIS_STATUS_NETWORK_CHANGE:
1135                 /* Only makes sense if carrier is present */
1136                 if (!rdev->link_state) {
1137                         rdev->link_state = true;
1138                         netif_carrier_off(net);
1139                         netif_tx_stop_all_queues(net);
1140                         event->event = RNDIS_STATUS_MEDIA_CONNECT;
1141                         spin_lock_irqsave(&ndev_ctx->lock, flags);
1142                         list_add(&event->list, &ndev_ctx->reconfig_events);
1143                         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1144                         reschedule = true;
1145                 }
1146                 break;
1147         }
1148
1149         rtnl_unlock();
1150
1151         if (notify)
1152                 netdev_notify_peers(net);
1153
1154         /* link_watch only sends one notification with current state per
1155          * second, handle next reconfig event in 2 seconds.
1156          */
1157         if (reschedule)
1158                 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1159
1160         return;
1161
1162 out_unlock:
1163         rtnl_unlock();
1164 }
1165
1166 static void netvsc_free_netdev(struct net_device *netdev)
1167 {
1168         struct net_device_context *net_device_ctx = netdev_priv(netdev);
1169
1170         free_percpu(net_device_ctx->tx_stats);
1171         free_percpu(net_device_ctx->rx_stats);
1172         free_netdev(netdev);
1173 }
1174
1175 static struct net_device *get_netvsc_bymac(const u8 *mac)
1176 {
1177         struct net_device *dev;
1178
1179         ASSERT_RTNL();
1180
1181         for_each_netdev(&init_net, dev) {
1182                 if (dev->netdev_ops != &device_ops)
1183                         continue;       /* not a netvsc device */
1184
1185                 if (ether_addr_equal(mac, dev->perm_addr))
1186                         return dev;
1187         }
1188
1189         return NULL;
1190 }
1191
1192 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
1193 {
1194         struct net_device *dev;
1195
1196         ASSERT_RTNL();
1197
1198         for_each_netdev(&init_net, dev) {
1199                 struct net_device_context *net_device_ctx;
1200
1201                 if (dev->netdev_ops != &device_ops)
1202                         continue;       /* not a netvsc device */
1203
1204                 net_device_ctx = netdev_priv(dev);
1205                 if (net_device_ctx->nvdev == NULL)
1206                         continue;       /* device is removed */
1207
1208                 if (rtnl_dereference(net_device_ctx->vf_netdev) == vf_netdev)
1209                         return dev;     /* a match */
1210         }
1211
1212         return NULL;
1213 }
1214
1215 static int netvsc_register_vf(struct net_device *vf_netdev)
1216 {
1217         struct net_device *ndev;
1218         struct net_device_context *net_device_ctx;
1219         struct netvsc_device *netvsc_dev;
1220
1221         if (vf_netdev->addr_len != ETH_ALEN)
1222                 return NOTIFY_DONE;
1223
1224         /*
1225          * We will use the MAC address to locate the synthetic interface to
1226          * associate with the VF interface. If we don't find a matching
1227          * synthetic interface, move on.
1228          */
1229         ndev = get_netvsc_bymac(vf_netdev->perm_addr);
1230         if (!ndev)
1231                 return NOTIFY_DONE;
1232
1233         net_device_ctx = netdev_priv(ndev);
1234         netvsc_dev = net_device_ctx->nvdev;
1235         if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
1236                 return NOTIFY_DONE;
1237
1238         netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
1239         /*
1240          * Take a reference on the module.
1241          */
1242         try_module_get(THIS_MODULE);
1243
1244         dev_hold(vf_netdev);
1245         rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
1246         return NOTIFY_OK;
1247 }
1248
1249 static int netvsc_vf_up(struct net_device *vf_netdev)
1250 {
1251         struct net_device *ndev;
1252         struct netvsc_device *netvsc_dev;
1253         struct net_device_context *net_device_ctx;
1254
1255         ndev = get_netvsc_byref(vf_netdev);
1256         if (!ndev)
1257                 return NOTIFY_DONE;
1258
1259         net_device_ctx = netdev_priv(ndev);
1260         netvsc_dev = net_device_ctx->nvdev;
1261
1262         netdev_info(ndev, "VF up: %s\n", vf_netdev->name);
1263
1264         /*
1265          * Open the device before switching data path.
1266          */
1267         rndis_filter_open(netvsc_dev);
1268
1269         /*
1270          * notify the host to switch the data path.
1271          */
1272         netvsc_switch_datapath(ndev, true);
1273         netdev_info(ndev, "Data path switched to VF: %s\n", vf_netdev->name);
1274
1275         netif_carrier_off(ndev);
1276
1277         /* Now notify peers through VF device. */
1278         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, vf_netdev);
1279
1280         return NOTIFY_OK;
1281 }
1282
1283 static int netvsc_vf_down(struct net_device *vf_netdev)
1284 {
1285         struct net_device *ndev;
1286         struct netvsc_device *netvsc_dev;
1287         struct net_device_context *net_device_ctx;
1288
1289         ndev = get_netvsc_byref(vf_netdev);
1290         if (!ndev)
1291                 return NOTIFY_DONE;
1292
1293         net_device_ctx = netdev_priv(ndev);
1294         netvsc_dev = net_device_ctx->nvdev;
1295
1296         netdev_info(ndev, "VF down: %s\n", vf_netdev->name);
1297         netvsc_switch_datapath(ndev, false);
1298         netdev_info(ndev, "Data path switched from VF: %s\n", vf_netdev->name);
1299         rndis_filter_close(netvsc_dev);
1300         netif_carrier_on(ndev);
1301
1302         /* Now notify peers through netvsc device. */
1303         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, ndev);
1304
1305         return NOTIFY_OK;
1306 }
1307
1308 static int netvsc_unregister_vf(struct net_device *vf_netdev)
1309 {
1310         struct net_device *ndev;
1311         struct netvsc_device *netvsc_dev;
1312         struct net_device_context *net_device_ctx;
1313
1314         ndev = get_netvsc_byref(vf_netdev);
1315         if (!ndev)
1316                 return NOTIFY_DONE;
1317
1318         net_device_ctx = netdev_priv(ndev);
1319         netvsc_dev = net_device_ctx->nvdev;
1320
1321         netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
1322
1323         RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
1324         dev_put(vf_netdev);
1325         module_put(THIS_MODULE);
1326         return NOTIFY_OK;
1327 }
1328
1329 static int netvsc_probe(struct hv_device *dev,
1330                         const struct hv_vmbus_device_id *dev_id)
1331 {
1332         struct net_device *net = NULL;
1333         struct net_device_context *net_device_ctx;
1334         struct netvsc_device_info device_info;
1335         struct netvsc_device *nvdev;
1336         int ret;
1337
1338         net = alloc_etherdev_mq(sizeof(struct net_device_context),
1339                                 num_online_cpus());
1340         if (!net)
1341                 return -ENOMEM;
1342
1343         netif_carrier_off(net);
1344
1345         netvsc_init_settings(net);
1346
1347         net_device_ctx = netdev_priv(net);
1348         net_device_ctx->device_ctx = dev;
1349         net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
1350         if (netif_msg_probe(net_device_ctx))
1351                 netdev_dbg(net, "netvsc msg_enable: %d\n",
1352                            net_device_ctx->msg_enable);
1353
1354         net_device_ctx->tx_stats = netdev_alloc_pcpu_stats(struct netvsc_stats);
1355         if (!net_device_ctx->tx_stats) {
1356                 free_netdev(net);
1357                 return -ENOMEM;
1358         }
1359         net_device_ctx->rx_stats = netdev_alloc_pcpu_stats(struct netvsc_stats);
1360         if (!net_device_ctx->rx_stats) {
1361                 free_percpu(net_device_ctx->tx_stats);
1362                 free_netdev(net);
1363                 return -ENOMEM;
1364         }
1365
1366         hv_set_drvdata(dev, net);
1367
1368         net_device_ctx->start_remove = false;
1369
1370         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
1371         INIT_WORK(&net_device_ctx->work, do_set_multicast);
1372
1373         spin_lock_init(&net_device_ctx->lock);
1374         INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
1375
1376         net->netdev_ops = &device_ops;
1377
1378         net->hw_features = NETVSC_HW_FEATURES;
1379         net->features = NETVSC_HW_FEATURES | NETIF_F_HW_VLAN_CTAG_TX;
1380
1381         net->ethtool_ops = &ethtool_ops;
1382         SET_NETDEV_DEV(net, &dev->device);
1383
1384         /* We always need headroom for rndis header */
1385         net->needed_headroom = RNDIS_AND_PPI_SIZE;
1386
1387         /* Notify the netvsc driver of the new device */
1388         memset(&device_info, 0, sizeof(device_info));
1389         device_info.ring_size = ring_size;
1390         device_info.max_num_vrss_chns = max_num_vrss_chns;
1391         ret = rndis_filter_device_add(dev, &device_info);
1392         if (ret != 0) {
1393                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
1394                 netvsc_free_netdev(net);
1395                 hv_set_drvdata(dev, NULL);
1396                 return ret;
1397         }
1398         memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
1399
1400         nvdev = net_device_ctx->nvdev;
1401         netif_set_real_num_tx_queues(net, nvdev->num_chn);
1402         netif_set_real_num_rx_queues(net, nvdev->num_chn);
1403
1404         ret = register_netdev(net);
1405         if (ret != 0) {
1406                 pr_err("Unable to register netdev.\n");
1407                 rndis_filter_device_remove(dev);
1408                 netvsc_free_netdev(net);
1409         }
1410
1411         return ret;
1412 }
1413
1414 static int netvsc_remove(struct hv_device *dev)
1415 {
1416         struct net_device *net;
1417         struct net_device_context *ndev_ctx;
1418         struct netvsc_device *net_device;
1419
1420         net = hv_get_drvdata(dev);
1421
1422         if (net == NULL) {
1423                 dev_err(&dev->device, "No net device to remove\n");
1424                 return 0;
1425         }
1426
1427         ndev_ctx = netdev_priv(net);
1428         net_device = ndev_ctx->nvdev;
1429
1430         /* Avoid racing with netvsc_change_mtu()/netvsc_set_channels()
1431          * removing the device.
1432          */
1433         rtnl_lock();
1434         ndev_ctx->start_remove = true;
1435         rtnl_unlock();
1436
1437         cancel_delayed_work_sync(&ndev_ctx->dwork);
1438         cancel_work_sync(&ndev_ctx->work);
1439
1440         /* Stop outbound asap */
1441         netif_tx_disable(net);
1442
1443         unregister_netdev(net);
1444
1445         /*
1446          * Call to the vsc driver to let it know that the device is being
1447          * removed
1448          */
1449         rndis_filter_device_remove(dev);
1450
1451         hv_set_drvdata(dev, NULL);
1452
1453         netvsc_free_netdev(net);
1454         return 0;
1455 }
1456
1457 static const struct hv_vmbus_device_id id_table[] = {
1458         /* Network guid */
1459         { HV_NIC_GUID, },
1460         { },
1461 };
1462
1463 MODULE_DEVICE_TABLE(vmbus, id_table);
1464
1465 /* The one and only one */
1466 static struct  hv_driver netvsc_drv = {
1467         .name = KBUILD_MODNAME,
1468         .id_table = id_table,
1469         .probe = netvsc_probe,
1470         .remove = netvsc_remove,
1471 };
1472
1473 /*
1474  * On Hyper-V, every VF interface is matched with a corresponding
1475  * synthetic interface. The synthetic interface is presented first
1476  * to the guest. When the corresponding VF instance is registered,
1477  * we will take care of switching the data path.
1478  */
1479 static int netvsc_netdev_event(struct notifier_block *this,
1480                                unsigned long event, void *ptr)
1481 {
1482         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
1483
1484         /* Skip our own events */
1485         if (event_dev->netdev_ops == &device_ops)
1486                 return NOTIFY_DONE;
1487
1488         /* Avoid non-Ethernet type devices */
1489         if (event_dev->type != ARPHRD_ETHER)
1490                 return NOTIFY_DONE;
1491
1492         /* Avoid Vlan dev with same MAC registering as VF */
1493         if (event_dev->priv_flags & IFF_802_1Q_VLAN)
1494                 return NOTIFY_DONE;
1495
1496         /* Avoid Bonding master dev with same MAC registering as VF */
1497         if ((event_dev->priv_flags & IFF_BONDING) &&
1498             (event_dev->flags & IFF_MASTER))
1499                 return NOTIFY_DONE;
1500
1501         switch (event) {
1502         case NETDEV_REGISTER:
1503                 return netvsc_register_vf(event_dev);
1504         case NETDEV_UNREGISTER:
1505                 return netvsc_unregister_vf(event_dev);
1506         case NETDEV_UP:
1507                 return netvsc_vf_up(event_dev);
1508         case NETDEV_DOWN:
1509                 return netvsc_vf_down(event_dev);
1510         default:
1511                 return NOTIFY_DONE;
1512         }
1513 }
1514
1515 static struct notifier_block netvsc_netdev_notifier = {
1516         .notifier_call = netvsc_netdev_event,
1517 };
1518
1519 static void __exit netvsc_drv_exit(void)
1520 {
1521         unregister_netdevice_notifier(&netvsc_netdev_notifier);
1522         vmbus_driver_unregister(&netvsc_drv);
1523 }
1524
1525 static int __init netvsc_drv_init(void)
1526 {
1527         int ret;
1528
1529         if (ring_size < RING_SIZE_MIN) {
1530                 ring_size = RING_SIZE_MIN;
1531                 pr_info("Increased ring_size to %d (min allowed)\n",
1532                         ring_size);
1533         }
1534         ret = vmbus_driver_register(&netvsc_drv);
1535
1536         if (ret)
1537                 return ret;
1538
1539         register_netdevice_notifier(&netvsc_netdev_notifier);
1540         return 0;
1541 }
1542
1543 MODULE_LICENSE("GPL");
1544 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
1545
1546 module_init(netvsc_drv_init);
1547 module_exit(netvsc_drv_exit);