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
  44 #define RING_SIZE_MIN 64
  45 #define LINKCHANGE_INT (2 * HZ)
  46 #define NETVSC_HW_FEATURES      (NETIF_F_RXCSUM | \
  47                                  NETIF_F_SG | \
  48                                  NETIF_F_TSO | \
  49                                  NETIF_F_TSO6 | \
  50                                  NETIF_F_HW_CSUM)
  51 static int ring_size = 128;
  52 module_param(ring_size, int, S_IRUGO);
  53 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
  54
  55 static int max_num_vrss_chns = 8;
  56
  57 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
  58                                 NETIF_MSG_LINK | NETIF_MSG_IFUP |
  59                                 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
  60                                 NETIF_MSG_TX_ERR;
  61
  62 static int debug = -1;
  63 module_param(debug, int, S_IRUGO);
  64 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  65
  66 static void do_set_multicast(struct work_struct *w)
  67 {
  68         struct net_device_context *ndevctx =
  69                 container_of(w, struct net_device_context, work);
  70         struct hv_device *device_obj = ndevctx->device_ctx;
  71         struct net_device *ndev = hv_get_drvdata(device_obj);
  72         struct netvsc_device *nvdev = ndevctx->nvdev;
  73         struct rndis_device *rdev;
  74
  75         if (!nvdev)
  76                 return;
  77
  78         rdev = nvdev->extension;
  79         if (rdev == NULL)
  80                 return;
  81
  82         if (ndev->flags & IFF_PROMISC)
  83                 rndis_filter_set_packet_filter(rdev,
  84                         NDIS_PACKET_TYPE_PROMISCUOUS);
  85         else
  86                 rndis_filter_set_packet_filter(rdev,
  87                         NDIS_PACKET_TYPE_BROADCAST |
  88                         NDIS_PACKET_TYPE_ALL_MULTICAST |
  89                         NDIS_PACKET_TYPE_DIRECTED);
  90 }
  91
  92 static void netvsc_set_multicast_list(struct net_device *net)
  93 {
  94         struct net_device_context *net_device_ctx = netdev_priv(net);
  95
  96         schedule_work(&net_device_ctx->work);
  97 }
  98
  99 static int netvsc_open(struct net_device *net)
 100 {
 101         struct net_device_context *net_device_ctx = netdev_priv(net);
 102         struct hv_device *device_obj = net_device_ctx->device_ctx;
 103         struct netvsc_device *nvdev = net_device_ctx->nvdev;
 104         struct rndis_device *rdev;
 105         int ret = 0;
 106
 107         netif_carrier_off(net);
 108
 109         /* Open up the device */
 110         ret = rndis_filter_open(device_obj);
 111         if (ret != 0) {
 112                 netdev_err(net, "unable to open device (ret %d).\n", ret);
 113                 return ret;
 114         }
 115
 116         netif_tx_wake_all_queues(net);
 117
 118         rdev = nvdev->extension;
 119         if (!rdev->link_state)
 120                 netif_carrier_on(net);
 121
 122         return ret;
 123 }
 124
 125 static int netvsc_close(struct net_device *net)
 126 {
 127         struct net_device_context *net_device_ctx = netdev_priv(net);
 128         struct hv_device *device_obj = net_device_ctx->device_ctx;
 129         struct netvsc_device *nvdev = net_device_ctx->nvdev;
 130         int ret;
 131         u32 aread, awrite, i, msec = 10, retry = 0, retry_max = 20;
 132         struct vmbus_channel *chn;
 133
 134         netif_tx_disable(net);
 135
 136         /* Make sure netvsc_set_multicast_list doesn't re-enable filter! */
 137         cancel_work_sync(&net_device_ctx->work);
 138         ret = rndis_filter_close(device_obj);
 139         if (ret != 0) {
 140                 netdev_err(net, "unable to close device (ret %d).\n", ret);
 141                 return ret;
 142         }
 143
 144         /* Ensure pending bytes in ring are read */
 145         while (true) {
 146                 aread = 0;
 147                 for (i = 0; i < nvdev->num_chn; i++) {
 148                         chn = nvdev->chn_table[i];
 149                         if (!chn)
 150                                 continue;
 151
 152                         hv_get_ringbuffer_availbytes(&chn->inbound, &aread,
 153                                                      &awrite);
 154
 155                         if (aread)
 156                                 break;
 157
 158                         hv_get_ringbuffer_availbytes(&chn->outbound, &aread,
 159                                                      &awrite);
 160
 161                         if (aread)
 162                                 break;
 163                 }
 164
 165                 retry++;
 166                 if (retry > retry_max || aread == 0)
 167                         break;
 168
 169                 msleep(msec);
 170
 171                 if (msec < 1000)
 172                         msec *= 2;
 173         }
 174
 175         if (aread) {
 176                 netdev_err(net, "Ring buffer not empty after closing rndis\n");
 177                 ret = -ETIMEDOUT;
 178         }
 179
 180         return ret;
 181 }
 182
 183 static void *init_ppi_data(struct rndis_message *msg, u32 ppi_size,
 184                                 int pkt_type)
 185 {
 186         struct rndis_packet *rndis_pkt;
 187         struct rndis_per_packet_info *ppi;
 188
 189         rndis_pkt = &msg->msg.pkt;
 190         rndis_pkt->data_offset += ppi_size;
 191
 192         ppi = (struct rndis_per_packet_info *)((void *)rndis_pkt +
 193                 rndis_pkt->per_pkt_info_offset + rndis_pkt->per_pkt_info_len);
 194
 195         ppi->size = ppi_size;
 196         ppi->type = pkt_type;
 197         ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
 198
 199         rndis_pkt->per_pkt_info_len += ppi_size;
 200
 201         return ppi;
 202 }
 203
 204 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
 205                         void *accel_priv, select_queue_fallback_t fallback)
 206 {
 207         struct net_device_context *net_device_ctx = netdev_priv(ndev);
 208         struct netvsc_device *nvsc_dev = net_device_ctx->nvdev;
 209         u32 hash;
 210         u16 q_idx = 0;
 211
 212         if (nvsc_dev == NULL || ndev->real_num_tx_queues <= 1)
 213                 return 0;
 214
 215         hash = skb_get_hash(skb);
 216         q_idx = nvsc_dev->send_table[hash % VRSS_SEND_TAB_SIZE] %
 217                 ndev->real_num_tx_queues;
 218
 219         if (!nvsc_dev->chn_table[q_idx])
 220                 q_idx = 0;
 221
 222         return q_idx;
 223 }
 224
 225 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
 226                         struct hv_page_buffer *pb)
 227 {
 228         int j = 0;
 229
 230         /* Deal with compund pages by ignoring unused part
 231          * of the page.
 232          */
 233         page += (offset >> PAGE_SHIFT);
 234         offset &= ~PAGE_MASK;
 235
 236         while (len > 0) {
 237                 unsigned long bytes;
 238
 239                 bytes = PAGE_SIZE - offset;
 240                 if (bytes > len)
 241                         bytes = len;
 242                 pb[j].pfn = page_to_pfn(page);
 243                 pb[j].offset = offset;
 244                 pb[j].len = bytes;
 245
 246                 offset += bytes;
 247                 len -= bytes;
 248
 249                 if (offset == PAGE_SIZE && len) {
 250                         page++;
 251                         offset = 0;
 252                         j++;
 253                 }
 254         }
 255
 256         return j + 1;
 257 }
 258
 259 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
 260                            struct hv_netvsc_packet *packet,
 261                            struct hv_page_buffer **page_buf)
 262 {
 263         struct hv_page_buffer *pb = *page_buf;
 264         u32 slots_used = 0;
 265         char *data = skb->data;
 266         int frags = skb_shinfo(skb)->nr_frags;
 267         int i;
 268
 269         /* The packet is laid out thus:
 270          * 1. hdr: RNDIS header and PPI
 271          * 2. skb linear data
 272          * 3. skb fragment data
 273          */
 274         if (hdr != NULL)
 275                 slots_used += fill_pg_buf(virt_to_page(hdr),
 276                                         offset_in_page(hdr),
 277                                         len, &pb[slots_used]);
 278
 279         packet->rmsg_size = len;
 280         packet->rmsg_pgcnt = slots_used;
 281
 282         slots_used += fill_pg_buf(virt_to_page(data),
 283                                 offset_in_page(data),
 284                                 skb_headlen(skb), &pb[slots_used]);
 285
 286         for (i = 0; i < frags; i++) {
 287                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 288
 289                 slots_used += fill_pg_buf(skb_frag_page(frag),
 290                                         frag->page_offset,
 291                                         skb_frag_size(frag), &pb[slots_used]);
 292         }
 293         return slots_used;
 294 }
 295
 296 static int count_skb_frag_slots(struct sk_buff *skb)
 297 {
 298         int i, frags = skb_shinfo(skb)->nr_frags;
 299         int pages = 0;
 300
 301         for (i = 0; i < frags; i++) {
 302                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 303                 unsigned long size = skb_frag_size(frag);
 304                 unsigned long offset = frag->page_offset;
 305
 306                 /* Skip unused frames from start of page */
 307                 offset &= ~PAGE_MASK;
 308                 pages += PFN_UP(offset + size);
 309         }
 310         return pages;
 311 }
 312
 313 static int netvsc_get_slots(struct sk_buff *skb)
 314 {
 315         char *data = skb->data;
 316         unsigned int offset = offset_in_page(data);
 317         unsigned int len = skb_headlen(skb);
 318         int slots;
 319         int frag_slots;
 320
 321         slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
 322         frag_slots = count_skb_frag_slots(skb);
 323         return slots + frag_slots;
 324 }
 325
 326 static u32 get_net_transport_info(struct sk_buff *skb, u32 *trans_off)
 327 {
 328         u32 ret_val = TRANSPORT_INFO_NOT_IP;
 329
 330         if ((eth_hdr(skb)->h_proto != htons(ETH_P_IP)) &&
 331                 (eth_hdr(skb)->h_proto != htons(ETH_P_IPV6))) {
 332                 goto not_ip;
 333         }
 334
 335         *trans_off = skb_transport_offset(skb);
 336
 337         if ((eth_hdr(skb)->h_proto == htons(ETH_P_IP))) {
 338                 struct iphdr *iphdr = ip_hdr(skb);
 339
 340                 if (iphdr->protocol == IPPROTO_TCP)
 341                         ret_val = TRANSPORT_INFO_IPV4_TCP;
 342                 else if (iphdr->protocol == IPPROTO_UDP)
 343                         ret_val = TRANSPORT_INFO_IPV4_UDP;
 344         } else {
 345                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
 346                         ret_val = TRANSPORT_INFO_IPV6_TCP;
 347                 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
 348                         ret_val = TRANSPORT_INFO_IPV6_UDP;
 349         }
 350
 351 not_ip:
 352         return ret_val;
 353 }
 354
 355 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
 356 {
 357         struct net_device_context *net_device_ctx = netdev_priv(net);
 358         struct hv_netvsc_packet *packet = NULL;
 359         int ret;
 360         unsigned int num_data_pgs;
 361         struct rndis_message *rndis_msg;
 362         struct rndis_packet *rndis_pkt;
 363         u32 rndis_msg_size;
 364         bool isvlan;
 365         bool linear = false;
 366         struct rndis_per_packet_info *ppi;
 367         struct ndis_tcp_ip_checksum_info *csum_info;
 368         struct ndis_tcp_lso_info *lso_info;
 369         int  hdr_offset;
 370         u32 net_trans_info;
 371         u32 hash;
 372         u32 skb_length;
 373         struct hv_page_buffer page_buf[MAX_PAGE_BUFFER_COUNT];
 374         struct hv_page_buffer *pb = page_buf;
 375         struct netvsc_stats *tx_stats = this_cpu_ptr(net_device_ctx->tx_stats);
 376
 377         /* We will atmost need two pages to describe the rndis
 378          * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
 379          * of pages in a single packet. If skb is scattered around
 380          * more pages we try linearizing it.
 381          */
 382
 383 check_size:
 384         skb_length = skb->len;
 385         num_data_pgs = netvsc_get_slots(skb) + 2;
 386         if (num_data_pgs > MAX_PAGE_BUFFER_COUNT && linear) {
 387                 net_alert_ratelimited("packet too big: %u pages (%u bytes)\n",
 388                                       num_data_pgs, skb->len);
 389                 ret = -EFAULT;
 390                 goto drop;
 391         } else if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
 392                 if (skb_linearize(skb)) {
 393                         net_alert_ratelimited("failed to linearize skb\n");
 394                         ret = -ENOMEM;
 395                         goto drop;
 396                 }
 397                 linear = true;
 398                 goto check_size;
 399         }
 400
 401         /*
 402          * Place the rndis header in the skb head room and
 403          * the skb->cb will be used for hv_netvsc_packet
 404          * structure.
 405          */
 406         ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
 407         if (ret) {
 408                 netdev_err(net, "unable to alloc hv_netvsc_packet\n");
 409                 ret = -ENOMEM;
 410                 goto drop;
 411         }
 412         /* Use the skb control buffer for building up the packet */
 413         BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
 414                         FIELD_SIZEOF(struct sk_buff, cb));
 415         packet = (struct hv_netvsc_packet *)skb->cb;
 416
 417
 418         packet->q_idx = skb_get_queue_mapping(skb);
 419
 420         packet->total_data_buflen = skb->len;
 421
 422         rndis_msg = (struct rndis_message *)skb->head;
 423
 424         memset(rndis_msg, 0, RNDIS_AND_PPI_SIZE);
 425
 426         isvlan = skb->vlan_tci & VLAN_TAG_PRESENT;
 427
 428         /* Add the rndis header */
 429         rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
 430         rndis_msg->msg_len = packet->total_data_buflen;
 431         rndis_pkt = &rndis_msg->msg.pkt;
 432         rndis_pkt->data_offset = sizeof(struct rndis_packet);
 433         rndis_pkt->data_len = packet->total_data_buflen;
 434         rndis_pkt->per_pkt_info_offset = sizeof(struct rndis_packet);
 435
 436         rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
 437
 438         hash = skb_get_hash_raw(skb);
 439         if (hash != 0 && net->real_num_tx_queues > 1) {
 440                 rndis_msg_size += NDIS_HASH_PPI_SIZE;
 441                 ppi = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
 442                                     NBL_HASH_VALUE);
 443                 *(u32 *)((void *)ppi + ppi->ppi_offset) = hash;
 444         }
 445
 446         if (isvlan) {
 447                 struct ndis_pkt_8021q_info *vlan;
 448
 449                 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
 450                 ppi = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
 451                                         IEEE_8021Q_INFO);
 452                 vlan = (struct ndis_pkt_8021q_info *)((void *)ppi +
 453                                                 ppi->ppi_offset);
 454                 vlan->vlanid = skb->vlan_tci & VLAN_VID_MASK;
 455                 vlan->pri = (skb->vlan_tci & VLAN_PRIO_MASK) >>
 456                                 VLAN_PRIO_SHIFT;
 457         }
 458
 459         net_trans_info = get_net_transport_info(skb, &hdr_offset);
 460         if (net_trans_info == TRANSPORT_INFO_NOT_IP)
 461                 goto do_send;
 462
 463         /*
 464          * Setup the sendside checksum offload only if this is not a
 465          * GSO packet.
 466          */
 467         if (skb_is_gso(skb))
 468                 goto do_lso;
 469
 470         if ((skb->ip_summed == CHECKSUM_NONE) ||
 471             (skb->ip_summed == CHECKSUM_UNNECESSARY))
 472                 goto do_send;
 473
 474         rndis_msg_size += NDIS_CSUM_PPI_SIZE;
 475         ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
 476                             TCPIP_CHKSUM_PKTINFO);
 477
 478         csum_info = (struct ndis_tcp_ip_checksum_info *)((void *)ppi +
 479                         ppi->ppi_offset);
 480
 481         if (net_trans_info & (INFO_IPV4 << 16))
 482                 csum_info->transmit.is_ipv4 = 1;
 483         else
 484                 csum_info->transmit.is_ipv6 = 1;
 485
 486         if (net_trans_info & INFO_TCP) {
 487                 csum_info->transmit.tcp_checksum = 1;
 488                 csum_info->transmit.tcp_header_offset = hdr_offset;
 489         } else if (net_trans_info & INFO_UDP) {
 490                 /* UDP checksum offload is not supported on ws2008r2.
 491                  * Furthermore, on ws2012 and ws2012r2, there are some
 492                  * issues with udp checksum offload from Linux guests.
 493                  * (these are host issues).
 494                  * For now compute the checksum here.
 495                  */
 496                 struct udphdr *uh;
 497                 u16 udp_len;
 498
 499                 ret = skb_cow_head(skb, 0);
 500                 if (ret)
 501                         goto drop;
 502
 503                 uh = udp_hdr(skb);
 504                 udp_len = ntohs(uh->len);
 505                 uh->check = 0;
 506                 uh->check = csum_tcpudp_magic(ip_hdr(skb)->saddr,
 507                                               ip_hdr(skb)->daddr,
 508                                               udp_len, IPPROTO_UDP,
 509                                               csum_partial(uh, udp_len, 0));
 510                 if (uh->check == 0)
 511                         uh->check = CSUM_MANGLED_0;
 512
 513                 csum_info->transmit.udp_checksum = 0;
 514         }
 515         goto do_send;
 516
 517 do_lso:
 518         rndis_msg_size += NDIS_LSO_PPI_SIZE;
 519         ppi = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
 520                             TCP_LARGESEND_PKTINFO);
 521
 522         lso_info = (struct ndis_tcp_lso_info *)((void *)ppi +
 523                         ppi->ppi_offset);
 524
 525         lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
 526         if (net_trans_info & (INFO_IPV4 << 16)) {
 527                 lso_info->lso_v2_transmit.ip_version =
 528                         NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
 529                 ip_hdr(skb)->tot_len = 0;
 530                 ip_hdr(skb)->check = 0;
 531                 tcp_hdr(skb)->check =
 532                 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
 533                                    ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 534         } else {
 535                 lso_info->lso_v2_transmit.ip_version =
 536                         NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
 537                 ipv6_hdr(skb)->payload_len = 0;
 538                 tcp_hdr(skb)->check =
 539                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 540                                 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 541         }
 542         lso_info->lso_v2_transmit.tcp_header_offset = hdr_offset;
 543         lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
 544
 545 do_send:
 546         /* Start filling in the page buffers with the rndis hdr */
 547         rndis_msg->msg_len += rndis_msg_size;
 548         packet->total_data_buflen = rndis_msg->msg_len;
 549         packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
 550                                                skb, packet, &pb);
 551
 552         /* timestamp packet in software */
 553         skb_tx_timestamp(skb);
 554         ret = netvsc_send(net_device_ctx->device_ctx, packet,
 555                           rndis_msg, &pb, skb);
 556
 557 drop:
 558         if (ret == 0) {
 559                 u64_stats_update_begin(&tx_stats->syncp);
 560                 tx_stats->packets++;
 561                 tx_stats->bytes += skb_length;
 562                 u64_stats_update_end(&tx_stats->syncp);
 563         } else {
 564                 if (ret != -EAGAIN) {
 565                         dev_kfree_skb_any(skb);
 566                         net->stats.tx_dropped++;
 567                 }
 568         }
 569
 570         return (ret == -EAGAIN) ? NETDEV_TX_BUSY : NETDEV_TX_OK;
 571 }
 572
 573 /*
 574  * netvsc_linkstatus_callback - Link up/down notification
 575  */
 576 void netvsc_linkstatus_callback(struct hv_device *device_obj,
 577                                 struct rndis_message *resp)
 578 {
 579         struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
 580         struct net_device *net;
 581         struct net_device_context *ndev_ctx;
 582         struct netvsc_reconfig *event;
 583         unsigned long flags;
 584
 585         /* Handle link change statuses only */
 586         if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
 587             indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
 588             indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
 589                 return;
 590
 591         net = hv_get_drvdata(device_obj);
 592
 593         if (!net || net->reg_state != NETREG_REGISTERED)
 594                 return;
 595
 596         ndev_ctx = netdev_priv(net);
 597
 598         event = kzalloc(sizeof(*event), GFP_ATOMIC);
 599         if (!event)
 600                 return;
 601         event->event = indicate->status;
 602
 603         spin_lock_irqsave(&ndev_ctx->lock, flags);
 604         list_add_tail(&event->list, &ndev_ctx->reconfig_events);
 605         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
 606
 607         schedule_delayed_work(&ndev_ctx->dwork, 0);
 608 }
 609
 610
 611 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
 612                                 struct hv_netvsc_packet *packet,
 613                                 struct ndis_tcp_ip_checksum_info *csum_info,
 614                                 void *data, u16 vlan_tci)
 615 {
 616         struct sk_buff *skb;
 617
 618         skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
 619         if (!skb)
 620                 return skb;
 621
 622         /*
 623          * Copy to skb. This copy is needed here since the memory pointed by
 624          * hv_netvsc_packet cannot be deallocated
 625          */
 626         memcpy(skb_put(skb, packet->total_data_buflen), data,
 627                packet->total_data_buflen);
 628
 629         skb->protocol = eth_type_trans(skb, net);
 630         if (csum_info) {
 631                 /* We only look at the IP checksum here.
 632                  * Should we be dropping the packet if checksum
 633                  * failed? How do we deal with other checksums - TCP/UDP?
 634                  */
 635                 if (csum_info->receive.ip_checksum_succeeded)
 636                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 637                 else
 638                         skb->ip_summed = CHECKSUM_NONE;
 639         }
 640
 641         if (vlan_tci & VLAN_TAG_PRESENT)
 642                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 643                                        vlan_tci);
 644
 645         return skb;
 646 }
 647
 648 /*
 649  * netvsc_recv_callback -  Callback when we receive a packet from the
 650  * "wire" on the specified device.
 651  */
 652 int netvsc_recv_callback(struct hv_device *device_obj,
 653                                 struct hv_netvsc_packet *packet,
 654                                 void **data,
 655                                 struct ndis_tcp_ip_checksum_info *csum_info,
 656                                 struct vmbus_channel *channel,
 657                                 u16 vlan_tci)
 658 {
 659         struct net_device *net = hv_get_drvdata(device_obj);
 660         struct net_device_context *net_device_ctx = netdev_priv(net);
 661         struct sk_buff *skb;
 662         struct sk_buff *vf_skb;
 663         struct netvsc_stats *rx_stats;
 664         struct netvsc_device *netvsc_dev = net_device_ctx->nvdev;
 665         u32 bytes_recvd = packet->total_data_buflen;
 666         int ret = 0;
 667
 668         if (!net || net->reg_state != NETREG_REGISTERED)
 669                 return NVSP_STAT_FAIL;
 670
 671         if (READ_ONCE(netvsc_dev->vf_inject)) {
 672                 atomic_inc(&netvsc_dev->vf_use_cnt);
 673                 if (!READ_ONCE(netvsc_dev->vf_inject)) {
 674                         /*
 675                          * We raced; just move on.
 676                          */
 677                         atomic_dec(&netvsc_dev->vf_use_cnt);
 678                         goto vf_injection_done;
 679                 }
 680
 681                 /*
 682                  * Inject this packet into the VF inerface.
 683                  * On Hyper-V, multicast and brodcast packets
 684                  * are only delivered on the synthetic interface
 685                  * (after subjecting these to policy filters on
 686                  * the host). Deliver these via the VF interface
 687                  * in the guest.
 688                  */
 689                 vf_skb = netvsc_alloc_recv_skb(netvsc_dev->vf_netdev, packet,
 690                                                csum_info, *data, vlan_tci);
 691                 if (vf_skb != NULL) {
 692                         ++netvsc_dev->vf_netdev->stats.rx_packets;
 693                         netvsc_dev->vf_netdev->stats.rx_bytes += bytes_recvd;
 694                         netif_receive_skb(vf_skb);
 695                 } else {
 696                         ++net->stats.rx_dropped;
 697                         ret = NVSP_STAT_FAIL;
 698                 }
 699                 atomic_dec(&netvsc_dev->vf_use_cnt);
 700                 return ret;
 701         }
 702
 703 vf_injection_done:
 704         net_device_ctx = netdev_priv(net);
 705         rx_stats = this_cpu_ptr(net_device_ctx->rx_stats);
 706
 707         /* Allocate a skb - TODO direct I/O to pages? */
 708         skb = netvsc_alloc_recv_skb(net, packet, csum_info, *data, vlan_tci);
 709         if (unlikely(!skb)) {
 710                 ++net->stats.rx_dropped;
 711                 return NVSP_STAT_FAIL;
 712         }
 713         skb_record_rx_queue(skb, channel->
 714                             offermsg.offer.sub_channel_index);
 715
 716         u64_stats_update_begin(&rx_stats->syncp);
 717         rx_stats->packets++;
 718         rx_stats->bytes += packet->total_data_buflen;
 719         u64_stats_update_end(&rx_stats->syncp);
 720
 721         /*
 722          * Pass the skb back up. Network stack will deallocate the skb when it
 723          * is done.
 724          * TODO - use NAPI?
 725          */
 726         netif_rx(skb);
 727
 728         return 0;
 729 }
 730
 731 static void netvsc_get_drvinfo(struct net_device *net,
 732                                struct ethtool_drvinfo *info)
 733 {
 734         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
 735         strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
 736 }
 737
 738 static void netvsc_get_channels(struct net_device *net,
 739                                 struct ethtool_channels *channel)
 740 {
 741         struct net_device_context *net_device_ctx = netdev_priv(net);
 742         struct netvsc_device *nvdev = net_device_ctx->nvdev;
 743
 744         if (nvdev) {
 745                 channel->max_combined   = nvdev->max_chn;
 746                 channel->combined_count = nvdev->num_chn;
 747         }
 748 }
 749
 750 static int netvsc_set_channels(struct net_device *net,
 751                                struct ethtool_channels *channels)
 752 {
 753         struct net_device_context *net_device_ctx = netdev_priv(net);
 754         struct hv_device *dev = net_device_ctx->device_ctx;
 755         struct netvsc_device *nvdev = net_device_ctx->nvdev;
 756         struct netvsc_device_info device_info;
 757         u32 num_chn;
 758         u32 max_chn;
 759         int ret = 0;
 760         bool recovering = false;
 761
 762         if (net_device_ctx->start_remove || !nvdev || nvdev->destroy)
 763                 return -ENODEV;
 764
 765         num_chn = nvdev->num_chn;
 766         max_chn = min_t(u32, nvdev->max_chn, num_online_cpus());
 767
 768         if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5) {
 769                 pr_info("vRSS unsupported before NVSP Version 5\n");
 770                 return -EINVAL;
 771         }
 772
 773         /* We do not support rx, tx, or other */
 774         if (!channels ||
 775             channels->rx_count ||
 776             channels->tx_count ||
 777             channels->other_count ||
 778             (channels->combined_count < 1))
 779                 return -EINVAL;
 780
 781         if (channels->combined_count > max_chn) {
 782                 pr_info("combined channels too high, using %d\n", max_chn);
 783                 channels->combined_count = max_chn;
 784         }
 785
 786         ret = netvsc_close(net);
 787         if (ret)
 788                 goto out;
 789
 790  do_set:
 791         net_device_ctx->start_remove = true;
 792         rndis_filter_device_remove(dev);
 793
 794         nvdev->num_chn = channels->combined_count;
 795
 796         memset(&device_info, 0, sizeof(device_info));
 797         device_info.num_chn = nvdev->num_chn; /* passed to RNDIS */
 798         device_info.ring_size = ring_size;
 799         device_info.max_num_vrss_chns = max_num_vrss_chns;
 800
 801         ret = rndis_filter_device_add(dev, &device_info);
 802         if (ret) {
 803                 if (recovering) {
 804                         netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
 805                         return ret;
 806                 }
 807                 goto recover;
 808         }
 809
 810         nvdev = net_device_ctx->nvdev;
 811
 812         ret = netif_set_real_num_tx_queues(net, nvdev->num_chn);
 813         if (ret) {
 814                 if (recovering) {
 815                         netdev_err(net, "could not set tx queue count (ret %d)\n", ret);
 816                         return ret;
 817                 }
 818                 goto recover;
 819         }
 820
 821         ret = netif_set_real_num_rx_queues(net, nvdev->num_chn);
 822         if (ret) {
 823                 if (recovering) {
 824                         netdev_err(net, "could not set rx queue count (ret %d)\n", ret);
 825                         return ret;
 826                 }
 827                 goto recover;
 828         }
 829
 830  out:
 831         netvsc_open(net);
 832         net_device_ctx->start_remove = false;
 833         /* We may have missed link change notifications */
 834         schedule_delayed_work(&net_device_ctx->dwork, 0);
 835
 836         return ret;
 837
 838  recover:
 839         /* If the above failed, we attempt to recover through the same
 840          * process but with the original number of channels.
 841          */
 842         netdev_err(net, "could not set channels, recovering\n");
 843         recovering = true;
 844         channels->combined_count = num_chn;
 845         goto do_set;
 846 }
 847
 848 static bool netvsc_validate_ethtool_ss_cmd(const struct ethtool_cmd *cmd)
 849 {
 850         struct ethtool_cmd diff1 = *cmd;
 851         struct ethtool_cmd diff2 = {};
 852
 853         ethtool_cmd_speed_set(&diff1, 0);
 854         diff1.duplex = 0;
 855         /* advertising and cmd are usually set */
 856         diff1.advertising = 0;
 857         diff1.cmd = 0;
 858         /* We set port to PORT_OTHER */
 859         diff2.port = PORT_OTHER;
 860
 861         return !memcmp(&diff1, &diff2, sizeof(diff1));
 862 }
 863
 864 static void netvsc_init_settings(struct net_device *dev)
 865 {
 866         struct net_device_context *ndc = netdev_priv(dev);
 867
 868         ndc->speed = SPEED_UNKNOWN;
 869         ndc->duplex = DUPLEX_UNKNOWN;
 870 }
 871
 872 static int netvsc_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
 873 {
 874         struct net_device_context *ndc = netdev_priv(dev);
 875
 876         ethtool_cmd_speed_set(cmd, ndc->speed);
 877         cmd->duplex = ndc->duplex;
 878         cmd->port = PORT_OTHER;
 879
 880         return 0;
 881 }
 882
 883 static int netvsc_set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
 884 {
 885         struct net_device_context *ndc = netdev_priv(dev);
 886         u32 speed;
 887
 888         speed = ethtool_cmd_speed(cmd);
 889         if (!ethtool_validate_speed(speed) ||
 890             !ethtool_validate_duplex(cmd->duplex) ||
 891             !netvsc_validate_ethtool_ss_cmd(cmd))
 892                 return -EINVAL;
 893
 894         ndc->speed = speed;
 895         ndc->duplex = cmd->duplex;
 896
 897         return 0;
 898 }
 899
 900 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
 901 {
 902         struct net_device_context *ndevctx = netdev_priv(ndev);
 903         struct netvsc_device *nvdev = ndevctx->nvdev;
 904         struct hv_device *hdev = ndevctx->device_ctx;
 905         struct netvsc_device_info device_info;
 906         int limit = ETH_DATA_LEN;
 907         u32 num_chn;
 908         int ret = 0;
 909
 910         if (ndevctx->start_remove || !nvdev || nvdev->destroy)
 911                 return -ENODEV;
 912
 913         if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
 914                 limit = NETVSC_MTU - ETH_HLEN;
 915
 916         if (mtu < NETVSC_MTU_MIN || mtu > limit)
 917                 return -EINVAL;
 918
 919         ret = netvsc_close(ndev);
 920         if (ret)
 921                 goto out;
 922
 923         num_chn = nvdev->num_chn;
 924
 925         ndevctx->start_remove = true;
 926         rndis_filter_device_remove(hdev);
 927
 928         ndev->mtu = mtu;
 929
 930         memset(&device_info, 0, sizeof(device_info));
 931         device_info.ring_size = ring_size;
 932         device_info.num_chn = num_chn;
 933         device_info.max_num_vrss_chns = max_num_vrss_chns;
 934         rndis_filter_device_add(hdev, &device_info);
 935
 936 out:
 937         netvsc_open(ndev);
 938         ndevctx->start_remove = false;
 939
 940         /* We may have missed link change notifications */
 941         schedule_delayed_work(&ndevctx->dwork, 0);
 942
 943         return ret;
 944 }
 945
 946 static struct rtnl_link_stats64 *netvsc_get_stats64(struct net_device *net,
 947                                                     struct rtnl_link_stats64 *t)
 948 {
 949         struct net_device_context *ndev_ctx = netdev_priv(net);
 950         int cpu;
 951
 952         for_each_possible_cpu(cpu) {
 953                 struct netvsc_stats *tx_stats = per_cpu_ptr(ndev_ctx->tx_stats,
 954                                                             cpu);
 955                 struct netvsc_stats *rx_stats = per_cpu_ptr(ndev_ctx->rx_stats,
 956                                                             cpu);
 957                 u64 tx_packets, tx_bytes, rx_packets, rx_bytes;
 958                 unsigned int start;
 959
 960                 do {
 961                         start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
 962                         tx_packets = tx_stats->packets;
 963                         tx_bytes = tx_stats->bytes;
 964                 } while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
 965
 966                 do {
 967                         start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
 968                         rx_packets = rx_stats->packets;
 969                         rx_bytes = rx_stats->bytes;
 970                 } while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
 971
 972                 t->tx_bytes     += tx_bytes;
 973                 t->tx_packets   += tx_packets;
 974                 t->rx_bytes     += rx_bytes;
 975                 t->rx_packets   += rx_packets;
 976         }
 977
 978         t->tx_dropped   = net->stats.tx_dropped;
 979         t->tx_errors    = net->stats.tx_dropped;
 980
 981         t->rx_dropped   = net->stats.rx_dropped;
 982         t->rx_errors    = net->stats.rx_errors;
 983
 984         return t;
 985 }
 986
 987 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
 988 {
 989         struct net_device_context *ndevctx = netdev_priv(ndev);
 990         struct hv_device *hdev =  ndevctx->device_ctx;
 991         struct sockaddr *addr = p;
 992         char save_adr[ETH_ALEN];
 993         unsigned char save_aatype;
 994         int err;
 995
 996         memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
 997         save_aatype = ndev->addr_assign_type;
 998
 999         err = eth_mac_addr(ndev, p);
1000         if (err != 0)
1001                 return err;
1002
1003         err = rndis_filter_set_device_mac(hdev, addr->sa_data);
1004         if (err != 0) {
1005                 /* roll back to saved MAC */
1006                 memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
1007                 ndev->addr_assign_type = save_aatype;
1008         }
1009
1010         return err;
1011 }
1012
1013 #ifdef CONFIG_NET_POLL_CONTROLLER
1014 static void netvsc_poll_controller(struct net_device *net)
1015 {
1016         /* As netvsc_start_xmit() works synchronous we don't have to
1017          * trigger anything here.
1018          */
1019 }
1020 #endif
1021
1022 static const struct ethtool_ops ethtool_ops = {
1023         .get_drvinfo    = netvsc_get_drvinfo,
1024         .get_link       = ethtool_op_get_link,
1025         .get_channels   = netvsc_get_channels,
1026         .set_channels   = netvsc_set_channels,
1027         .get_ts_info    = ethtool_op_get_ts_info,
1028         .get_settings   = netvsc_get_settings,
1029         .set_settings   = netvsc_set_settings,
1030 };
1031
1032 static const struct net_device_ops device_ops = {
1033         .ndo_open =                     netvsc_open,
1034         .ndo_stop =                     netvsc_close,
1035         .ndo_start_xmit =               netvsc_start_xmit,
1036         .ndo_set_rx_mode =              netvsc_set_multicast_list,
1037         .ndo_change_mtu =               netvsc_change_mtu,
1038         .ndo_validate_addr =            eth_validate_addr,
1039         .ndo_set_mac_address =          netvsc_set_mac_addr,
1040         .ndo_select_queue =             netvsc_select_queue,
1041         .ndo_get_stats64 =              netvsc_get_stats64,
1042 #ifdef CONFIG_NET_POLL_CONTROLLER
1043         .ndo_poll_controller =          netvsc_poll_controller,
1044 #endif
1045 };
1046
1047 /*
1048  * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
1049  * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
1050  * present send GARP packet to network peers with netif_notify_peers().
1051  */
1052 static void netvsc_link_change(struct work_struct *w)
1053 {
1054         struct net_device_context *ndev_ctx =
1055                 container_of(w, struct net_device_context, dwork.work);
1056         struct hv_device *device_obj = ndev_ctx->device_ctx;
1057         struct net_device *net = hv_get_drvdata(device_obj);
1058         struct netvsc_device *net_device;
1059         struct rndis_device *rdev;
1060         struct netvsc_reconfig *event = NULL;
1061         bool notify = false, reschedule = false;
1062         unsigned long flags, next_reconfig, delay;
1063
1064         rtnl_lock();
1065         if (ndev_ctx->start_remove)
1066                 goto out_unlock;
1067
1068         net_device = ndev_ctx->nvdev;
1069         rdev = net_device->extension;
1070
1071         next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
1072         if (time_is_after_jiffies(next_reconfig)) {
1073                 /* link_watch only sends one notification with current state
1074                  * per second, avoid doing reconfig more frequently. Handle
1075                  * wrap around.
1076                  */
1077                 delay = next_reconfig - jiffies;
1078                 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
1079                 schedule_delayed_work(&ndev_ctx->dwork, delay);
1080                 goto out_unlock;
1081         }
1082         ndev_ctx->last_reconfig = jiffies;
1083
1084         spin_lock_irqsave(&ndev_ctx->lock, flags);
1085         if (!list_empty(&ndev_ctx->reconfig_events)) {
1086                 event = list_first_entry(&ndev_ctx->reconfig_events,
1087                                          struct netvsc_reconfig, list);
1088                 list_del(&event->list);
1089                 reschedule = !list_empty(&ndev_ctx->reconfig_events);
1090         }
1091         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1092
1093         if (!event)
1094                 goto out_unlock;
1095
1096         switch (event->event) {
1097                 /* Only the following events are possible due to the check in
1098                  * netvsc_linkstatus_callback()
1099                  */
1100         case RNDIS_STATUS_MEDIA_CONNECT:
1101                 if (rdev->link_state) {
1102                         rdev->link_state = false;
1103                         netif_carrier_on(net);
1104                         netif_tx_wake_all_queues(net);
1105                 } else {
1106                         notify = true;
1107                 }
1108                 kfree(event);
1109                 break;
1110         case RNDIS_STATUS_MEDIA_DISCONNECT:
1111                 if (!rdev->link_state) {
1112                         rdev->link_state = true;
1113                         netif_carrier_off(net);
1114                         netif_tx_stop_all_queues(net);
1115                 }
1116                 kfree(event);
1117                 break;
1118         case RNDIS_STATUS_NETWORK_CHANGE:
1119                 /* Only makes sense if carrier is present */
1120                 if (!rdev->link_state) {
1121                         rdev->link_state = true;
1122                         netif_carrier_off(net);
1123                         netif_tx_stop_all_queues(net);
1124                         event->event = RNDIS_STATUS_MEDIA_CONNECT;
1125                         spin_lock_irqsave(&ndev_ctx->lock, flags);
1126                         list_add(&event->list, &ndev_ctx->reconfig_events);
1127                         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1128                         reschedule = true;
1129                 }
1130                 break;
1131         }
1132
1133         rtnl_unlock();
1134
1135         if (notify)
1136                 netdev_notify_peers(net);
1137
1138         /* link_watch only sends one notification with current state per
1139          * second, handle next reconfig event in 2 seconds.
1140          */
1141         if (reschedule)
1142                 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1143
1144         return;
1145
1146 out_unlock:
1147         rtnl_unlock();
1148 }
1149
1150 static void netvsc_free_netdev(struct net_device *netdev)
1151 {
1152         struct net_device_context *net_device_ctx = netdev_priv(netdev);
1153
1154         free_percpu(net_device_ctx->tx_stats);
1155         free_percpu(net_device_ctx->rx_stats);
1156         free_netdev(netdev);
1157 }
1158
1159 static void netvsc_notify_peers(struct work_struct *wrk)
1160 {
1161         struct garp_wrk *gwrk;
1162
1163         gwrk = container_of(wrk, struct garp_wrk, dwrk);
1164
1165         netdev_notify_peers(gwrk->netdev);
1166
1167         atomic_dec(&gwrk->netvsc_dev->vf_use_cnt);
1168 }
1169
1170 static struct net_device *get_netvsc_net_device(char *mac)
1171 {
1172         struct net_device *dev, *found = NULL;
1173         int rtnl_locked;
1174
1175         rtnl_locked = rtnl_trylock();
1176
1177         for_each_netdev(&init_net, dev) {
1178                 if (memcmp(dev->dev_addr, mac, ETH_ALEN) == 0) {
1179                         if (dev->netdev_ops != &device_ops)
1180                                 continue;
1181                         found = dev;
1182                         break;
1183                 }
1184         }
1185         if (rtnl_locked)
1186                 rtnl_unlock();
1187
1188         return found;
1189 }
1190
1191 static int netvsc_register_vf(struct net_device *vf_netdev)
1192 {
1193         struct net_device *ndev;
1194         struct net_device_context *net_device_ctx;
1195         struct netvsc_device *netvsc_dev;
1196         const struct ethtool_ops *eth_ops = vf_netdev->ethtool_ops;
1197
1198         if (eth_ops == NULL || eth_ops == &ethtool_ops)
1199                 return NOTIFY_DONE;
1200
1201         /*
1202          * We will use the MAC address to locate the synthetic interface to
1203          * associate with the VF interface. If we don't find a matching
1204          * synthetic interface, move on.
1205          */
1206         ndev = get_netvsc_net_device(vf_netdev->dev_addr);
1207         if (!ndev)
1208                 return NOTIFY_DONE;
1209
1210         net_device_ctx = netdev_priv(ndev);
1211         netvsc_dev = net_device_ctx->nvdev;
1212         if (netvsc_dev == NULL)
1213                 return NOTIFY_DONE;
1214
1215         netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
1216         /*
1217          * Take a reference on the module.
1218          */
1219         try_module_get(THIS_MODULE);
1220         netvsc_dev->vf_netdev = vf_netdev;
1221         return NOTIFY_OK;
1222 }
1223
1224
1225 static int netvsc_vf_up(struct net_device *vf_netdev)
1226 {
1227         struct net_device *ndev;
1228         struct netvsc_device *netvsc_dev;
1229         const struct ethtool_ops *eth_ops = vf_netdev->ethtool_ops;
1230         struct net_device_context *net_device_ctx;
1231
1232         if (eth_ops == &ethtool_ops)
1233                 return NOTIFY_DONE;
1234
1235         ndev = get_netvsc_net_device(vf_netdev->dev_addr);
1236         if (!ndev)
1237                 return NOTIFY_DONE;
1238
1239         net_device_ctx = netdev_priv(ndev);
1240         netvsc_dev = net_device_ctx->nvdev;
1241
1242         if ((netvsc_dev == NULL) || (netvsc_dev->vf_netdev == NULL))
1243                 return NOTIFY_DONE;
1244
1245         netdev_info(ndev, "VF up: %s\n", vf_netdev->name);
1246         netvsc_dev->vf_inject = true;
1247
1248         /*
1249          * Open the device before switching data path.
1250          */
1251         rndis_filter_open(net_device_ctx->device_ctx);
1252
1253         /*
1254          * notify the host to switch the data path.
1255          */
1256         netvsc_switch_datapath(ndev, true);
1257         netdev_info(ndev, "Data path switched to VF: %s\n", vf_netdev->name);
1258
1259         netif_carrier_off(ndev);
1260
1261         /*
1262          * Now notify peers. We are scheduling work to
1263          * notify peers; take a reference to prevent
1264          * the VF interface from vanishing.
1265          */
1266         atomic_inc(&netvsc_dev->vf_use_cnt);
1267         net_device_ctx->gwrk.netdev = vf_netdev;
1268         net_device_ctx->gwrk.netvsc_dev = netvsc_dev;
1269         schedule_work(&net_device_ctx->gwrk.dwrk);
1270
1271         return NOTIFY_OK;
1272 }
1273
1274
1275 static int netvsc_vf_down(struct net_device *vf_netdev)
1276 {
1277         struct net_device *ndev;
1278         struct netvsc_device *netvsc_dev;
1279         struct net_device_context *net_device_ctx;
1280         const struct ethtool_ops *eth_ops = vf_netdev->ethtool_ops;
1281
1282         if (eth_ops == &ethtool_ops)
1283                 return NOTIFY_DONE;
1284
1285         ndev = get_netvsc_net_device(vf_netdev->dev_addr);
1286         if (!ndev)
1287                 return NOTIFY_DONE;
1288
1289         net_device_ctx = netdev_priv(ndev);
1290         netvsc_dev = net_device_ctx->nvdev;
1291
1292         if ((netvsc_dev == NULL) || (netvsc_dev->vf_netdev == NULL))
1293                 return NOTIFY_DONE;
1294
1295         netdev_info(ndev, "VF down: %s\n", vf_netdev->name);
1296         netvsc_dev->vf_inject = false;
1297         /*
1298          * Wait for currently active users to
1299          * drain out.
1300          */
1301
1302         while (atomic_read(&netvsc_dev->vf_use_cnt) != 0)
1303                 udelay(50);
1304         netvsc_switch_datapath(ndev, false);
1305         netdev_info(ndev, "Data path switched from VF: %s\n", vf_netdev->name);
1306         rndis_filter_close(net_device_ctx->device_ctx);
1307         netif_carrier_on(ndev);
1308         /*
1309          * Notify peers.
1310          */
1311         atomic_inc(&netvsc_dev->vf_use_cnt);
1312         net_device_ctx->gwrk.netdev = ndev;
1313         net_device_ctx->gwrk.netvsc_dev = netvsc_dev;
1314         schedule_work(&net_device_ctx->gwrk.dwrk);
1315
1316         return NOTIFY_OK;
1317 }
1318
1319
1320 static int netvsc_unregister_vf(struct net_device *vf_netdev)
1321 {
1322         struct net_device *ndev;
1323         struct netvsc_device *netvsc_dev;
1324         const struct ethtool_ops *eth_ops = vf_netdev->ethtool_ops;
1325         struct net_device_context *net_device_ctx;
1326
1327         if (eth_ops == &ethtool_ops)
1328                 return NOTIFY_DONE;
1329
1330         ndev = get_netvsc_net_device(vf_netdev->dev_addr);
1331         if (!ndev)
1332                 return NOTIFY_DONE;
1333
1334         net_device_ctx = netdev_priv(ndev);
1335         netvsc_dev = net_device_ctx->nvdev;
1336         if (netvsc_dev == NULL)
1337                 return NOTIFY_DONE;
1338         netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
1339
1340         netvsc_dev->vf_netdev = NULL;
1341         module_put(THIS_MODULE);
1342         return NOTIFY_OK;
1343 }
1344
1345 static int netvsc_probe(struct hv_device *dev,
1346                         const struct hv_vmbus_device_id *dev_id)
1347 {
1348         struct net_device *net = NULL;
1349         struct net_device_context *net_device_ctx;
1350         struct netvsc_device_info device_info;
1351         struct netvsc_device *nvdev;
1352         int ret;
1353
1354         net = alloc_etherdev_mq(sizeof(struct net_device_context),
1355                                 num_online_cpus());
1356         if (!net)
1357                 return -ENOMEM;
1358
1359         netif_carrier_off(net);
1360
1361         net_device_ctx = netdev_priv(net);
1362         net_device_ctx->device_ctx = dev;
1363         net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
1364         if (netif_msg_probe(net_device_ctx))
1365                 netdev_dbg(net, "netvsc msg_enable: %d\n",
1366                            net_device_ctx->msg_enable);
1367
1368         net_device_ctx->tx_stats = netdev_alloc_pcpu_stats(struct netvsc_stats);
1369         if (!net_device_ctx->tx_stats) {
1370                 free_netdev(net);
1371                 return -ENOMEM;
1372         }
1373         net_device_ctx->rx_stats = netdev_alloc_pcpu_stats(struct netvsc_stats);
1374         if (!net_device_ctx->rx_stats) {
1375                 free_percpu(net_device_ctx->tx_stats);
1376                 free_netdev(net);
1377                 return -ENOMEM;
1378         }
1379
1380         hv_set_drvdata(dev, net);
1381
1382         net_device_ctx->start_remove = false;
1383
1384         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
1385         INIT_WORK(&net_device_ctx->work, do_set_multicast);
1386         INIT_WORK(&net_device_ctx->gwrk.dwrk, netvsc_notify_peers);
1387
1388         spin_lock_init(&net_device_ctx->lock);
1389         INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
1390
1391         net->netdev_ops = &device_ops;
1392
1393         net->hw_features = NETVSC_HW_FEATURES;
1394         net->features = NETVSC_HW_FEATURES | NETIF_F_HW_VLAN_CTAG_TX;
1395
1396         net->ethtool_ops = &ethtool_ops;
1397         SET_NETDEV_DEV(net, &dev->device);
1398
1399         /* We always need headroom for rndis header */
1400         net->needed_headroom = RNDIS_AND_PPI_SIZE;
1401
1402         /* Notify the netvsc driver of the new device */
1403         memset(&device_info, 0, sizeof(device_info));
1404         device_info.ring_size = ring_size;
1405         device_info.max_num_vrss_chns = max_num_vrss_chns;
1406         ret = rndis_filter_device_add(dev, &device_info);
1407         if (ret != 0) {
1408                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
1409                 netvsc_free_netdev(net);
1410                 hv_set_drvdata(dev, NULL);
1411                 return ret;
1412         }
1413         memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
1414
1415         nvdev = net_device_ctx->nvdev;
1416         netif_set_real_num_tx_queues(net, nvdev->num_chn);
1417         netif_set_real_num_rx_queues(net, nvdev->num_chn);
1418
1419         netvsc_init_settings(net);
1420
1421         ret = register_netdev(net);
1422         if (ret != 0) {
1423                 pr_err("Unable to register netdev.\n");
1424                 rndis_filter_device_remove(dev);
1425                 netvsc_free_netdev(net);
1426         }
1427
1428         return ret;
1429 }
1430
1431 static int netvsc_remove(struct hv_device *dev)
1432 {
1433         struct net_device *net;
1434         struct net_device_context *ndev_ctx;
1435         struct netvsc_device *net_device;
1436
1437         net = hv_get_drvdata(dev);
1438
1439         if (net == NULL) {
1440                 dev_err(&dev->device, "No net device to remove\n");
1441                 return 0;
1442         }
1443
1444
1445         ndev_ctx = netdev_priv(net);
1446         net_device = ndev_ctx->nvdev;
1447
1448         /* Avoid racing with netvsc_change_mtu()/netvsc_set_channels()
1449          * removing the device.
1450          */
1451         rtnl_lock();
1452         ndev_ctx->start_remove = true;
1453         rtnl_unlock();
1454
1455         cancel_delayed_work_sync(&ndev_ctx->dwork);
1456         cancel_work_sync(&ndev_ctx->work);
1457
1458         /* Stop outbound asap */
1459         netif_tx_disable(net);
1460
1461         unregister_netdev(net);
1462
1463         /*
1464          * Call to the vsc driver to let it know that the device is being
1465          * removed
1466          */
1467         rndis_filter_device_remove(dev);
1468
1469         hv_set_drvdata(dev, NULL);
1470
1471         netvsc_free_netdev(net);
1472         return 0;
1473 }
1474
1475 static const struct hv_vmbus_device_id id_table[] = {
1476         /* Network guid */
1477         { HV_NIC_GUID, },
1478         { },
1479 };
1480
1481 MODULE_DEVICE_TABLE(vmbus, id_table);
1482
1483 /* The one and only one */
1484 static struct  hv_driver netvsc_drv = {
1485         .name = KBUILD_MODNAME,
1486         .id_table = id_table,
1487         .probe = netvsc_probe,
1488         .remove = netvsc_remove,
1489 };
1490
1491
1492 /*
1493  * On Hyper-V, every VF interface is matched with a corresponding
1494  * synthetic interface. The synthetic interface is presented first
1495  * to the guest. When the corresponding VF instance is registered,
1496  * we will take care of switching the data path.
1497  */
1498 static int netvsc_netdev_event(struct notifier_block *this,
1499                                unsigned long event, void *ptr)
1500 {
1501         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
1502
1503         switch (event) {
1504         case NETDEV_REGISTER:
1505                 return netvsc_register_vf(event_dev);
1506         case NETDEV_UNREGISTER:
1507                 return netvsc_unregister_vf(event_dev);
1508         case NETDEV_UP:
1509                 return netvsc_vf_up(event_dev);
1510         case NETDEV_DOWN:
1511                 return netvsc_vf_down(event_dev);
1512         default:
1513                 return NOTIFY_DONE;
1514         }
1515 }
1516
1517 static struct notifier_block netvsc_netdev_notifier = {
1518         .notifier_call = netvsc_netdev_event,
1519 };
1520
1521 static void __exit netvsc_drv_exit(void)
1522 {
1523         unregister_netdevice_notifier(&netvsc_netdev_notifier);
1524         vmbus_driver_unregister(&netvsc_drv);
1525 }
1526
1527 static int __init netvsc_drv_init(void)
1528 {
1529         int ret;
1530
1531         if (ring_size < RING_SIZE_MIN) {
1532                 ring_size = RING_SIZE_MIN;
1533                 pr_info("Increased ring_size to %d (min allowed)\n",
1534                         ring_size);
1535         }
1536         ret = vmbus_driver_register(&netvsc_drv);
1537
1538         if (ret)
1539                 return ret;
1540
1541         register_netdevice_notifier(&netvsc_netdev_notifier);
1542         return 0;
1543 }
1544
1545 MODULE_LICENSE("GPL");
1546 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
1547
1548 module_init(netvsc_drv_init);
1549 module_exit(netvsc_drv_exit);