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 netvsc_device *nvdev = net_device_to_netvsc_device(net);
 102         struct rndis_device *rdev;
 103         int ret = 0;
 104
 105         netif_carrier_off(net);
 106
 107         /* Open up the device */
 108         ret = rndis_filter_open(nvdev);
 109         if (ret != 0) {
 110                 netdev_err(net, "unable to open device (ret %d).\n", ret);
 111                 return ret;
 112         }
 113
 114         netif_tx_wake_all_queues(net);
 115
 116         rdev = nvdev->extension;
 117         if (!rdev->link_state)
 118                 netif_carrier_on(net);
 119
 120         return ret;
 121 }
 122
 123 static int netvsc_close(struct net_device *net)
 124 {
 125         struct net_device_context *net_device_ctx = netdev_priv(net);
 126         struct netvsc_device *nvdev = net_device_ctx->nvdev;
 127         int ret;
 128         u32 aread, awrite, i, msec = 10, retry = 0, retry_max = 20;
 129         struct vmbus_channel *chn;
 130
 131         netif_tx_disable(net);
 132
 133         /* Make sure netvsc_set_multicast_list doesn't re-enable filter! */
 134         cancel_work_sync(&net_device_ctx->work);
 135         ret = rndis_filter_close(nvdev);
 136         if (ret != 0) {
 137                 netdev_err(net, "unable to close device (ret %d).\n", ret);
 138                 return ret;
 139         }
 140
 141         /* Ensure pending bytes in ring are read */
 142         while (true) {
 143                 aread = 0;
 144                 for (i = 0; i < nvdev->num_chn; i++) {
 145                         chn = nvdev->chn_table[i];
 146                         if (!chn)
 147                                 continue;
 148
 149                         hv_get_ringbuffer_availbytes(&chn->inbound, &aread,
 150                                                      &awrite);
 151
 152                         if (aread)
 153                                 break;
 154
 155                         hv_get_ringbuffer_availbytes(&chn->outbound, &aread,
 156                                                      &awrite);
 157
 158                         if (aread)
 159                                 break;
 160                 }
 161
 162                 retry++;
 163                 if (retry > retry_max || aread == 0)
 164                         break;
 165
 166                 msleep(msec);
 167
 168                 if (msec < 1000)
 169                         msec *= 2;
 170         }
 171
 172         if (aread) {
 173                 netdev_err(net, "Ring buffer not empty after closing rndis\n");
 174                 ret = -ETIMEDOUT;
 175         }
 176
 177         return ret;
 178 }
 179
 180 static void *init_ppi_data(struct rndis_message *msg, u32 ppi_size,
 181                                 int pkt_type)
 182 {
 183         struct rndis_packet *rndis_pkt;
 184         struct rndis_per_packet_info *ppi;
 185
 186         rndis_pkt = &msg->msg.pkt;
 187         rndis_pkt->data_offset += ppi_size;
 188
 189         ppi = (struct rndis_per_packet_info *)((void *)rndis_pkt +
 190                 rndis_pkt->per_pkt_info_offset + rndis_pkt->per_pkt_info_len);
 191
 192         ppi->size = ppi_size;
 193         ppi->type = pkt_type;
 194         ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
 195
 196         rndis_pkt->per_pkt_info_len += ppi_size;
 197
 198         return ppi;
 199 }
 200
 201 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
 202                         void *accel_priv, select_queue_fallback_t fallback)
 203 {
 204         struct net_device_context *net_device_ctx = netdev_priv(ndev);
 205         struct netvsc_device *nvsc_dev = net_device_ctx->nvdev;
 206         u32 hash;
 207         u16 q_idx = 0;
 208
 209         if (nvsc_dev == NULL || ndev->real_num_tx_queues <= 1)
 210                 return 0;
 211
 212         hash = skb_get_hash(skb);
 213         q_idx = nvsc_dev->send_table[hash % VRSS_SEND_TAB_SIZE] %
 214                 ndev->real_num_tx_queues;
 215
 216         if (!nvsc_dev->chn_table[q_idx])
 217                 q_idx = 0;
 218
 219         return q_idx;
 220 }
 221
 222 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
 223                         struct hv_page_buffer *pb)
 224 {
 225         int j = 0;
 226
 227         /* Deal with compund pages by ignoring unused part
 228          * of the page.
 229          */
 230         page += (offset >> PAGE_SHIFT);
 231         offset &= ~PAGE_MASK;
 232
 233         while (len > 0) {
 234                 unsigned long bytes;
 235
 236                 bytes = PAGE_SIZE - offset;
 237                 if (bytes > len)
 238                         bytes = len;
 239                 pb[j].pfn = page_to_pfn(page);
 240                 pb[j].offset = offset;
 241                 pb[j].len = bytes;
 242
 243                 offset += bytes;
 244                 len -= bytes;
 245
 246                 if (offset == PAGE_SIZE && len) {
 247                         page++;
 248                         offset = 0;
 249                         j++;
 250                 }
 251         }
 252
 253         return j + 1;
 254 }
 255
 256 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
 257                            struct hv_netvsc_packet *packet,
 258                            struct hv_page_buffer **page_buf)
 259 {
 260         struct hv_page_buffer *pb = *page_buf;
 261         u32 slots_used = 0;
 262         char *data = skb->data;
 263         int frags = skb_shinfo(skb)->nr_frags;
 264         int i;
 265
 266         /* The packet is laid out thus:
 267          * 1. hdr: RNDIS header and PPI
 268          * 2. skb linear data
 269          * 3. skb fragment data
 270          */
 271         if (hdr != NULL)
 272                 slots_used += fill_pg_buf(virt_to_page(hdr),
 273                                         offset_in_page(hdr),
 274                                         len, &pb[slots_used]);
 275
 276         packet->rmsg_size = len;
 277         packet->rmsg_pgcnt = slots_used;
 278
 279         slots_used += fill_pg_buf(virt_to_page(data),
 280                                 offset_in_page(data),
 281                                 skb_headlen(skb), &pb[slots_used]);
 282
 283         for (i = 0; i < frags; i++) {
 284                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 285
 286                 slots_used += fill_pg_buf(skb_frag_page(frag),
 287                                         frag->page_offset,
 288                                         skb_frag_size(frag), &pb[slots_used]);
 289         }
 290         return slots_used;
 291 }
 292
 293 static int count_skb_frag_slots(struct sk_buff *skb)
 294 {
 295         int i, frags = skb_shinfo(skb)->nr_frags;
 296         int pages = 0;
 297
 298         for (i = 0; i < frags; i++) {
 299                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 300                 unsigned long size = skb_frag_size(frag);
 301                 unsigned long offset = frag->page_offset;
 302
 303                 /* Skip unused frames from start of page */
 304                 offset &= ~PAGE_MASK;
 305                 pages += PFN_UP(offset + size);
 306         }
 307         return pages;
 308 }
 309
 310 static int netvsc_get_slots(struct sk_buff *skb)
 311 {
 312         char *data = skb->data;
 313         unsigned int offset = offset_in_page(data);
 314         unsigned int len = skb_headlen(skb);
 315         int slots;
 316         int frag_slots;
 317
 318         slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
 319         frag_slots = count_skb_frag_slots(skb);
 320         return slots + frag_slots;
 321 }
 322
 323 static u32 get_net_transport_info(struct sk_buff *skb, u32 *trans_off)
 324 {
 325         u32 ret_val = TRANSPORT_INFO_NOT_IP;
 326
 327         if ((eth_hdr(skb)->h_proto != htons(ETH_P_IP)) &&
 328                 (eth_hdr(skb)->h_proto != htons(ETH_P_IPV6))) {
 329                 goto not_ip;
 330         }
 331
 332         *trans_off = skb_transport_offset(skb);
 333
 334         if ((eth_hdr(skb)->h_proto == htons(ETH_P_IP))) {
 335                 struct iphdr *iphdr = ip_hdr(skb);
 336
 337                 if (iphdr->protocol == IPPROTO_TCP)
 338                         ret_val = TRANSPORT_INFO_IPV4_TCP;
 339                 else if (iphdr->protocol == IPPROTO_UDP)
 340                         ret_val = TRANSPORT_INFO_IPV4_UDP;
 341         } else {
 342                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
 343                         ret_val = TRANSPORT_INFO_IPV6_TCP;
 344                 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
 345                         ret_val = TRANSPORT_INFO_IPV6_UDP;
 346         }
 347
 348 not_ip:
 349         return ret_val;
 350 }
 351
 352 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
 353 {
 354         struct net_device_context *net_device_ctx = netdev_priv(net);
 355         struct hv_netvsc_packet *packet = NULL;
 356         int ret;
 357         unsigned int num_data_pgs;
 358         struct rndis_message *rndis_msg;
 359         struct rndis_packet *rndis_pkt;
 360         u32 rndis_msg_size;
 361         bool isvlan;
 362         bool linear = false;
 363         struct rndis_per_packet_info *ppi;
 364         struct ndis_tcp_ip_checksum_info *csum_info;
 365         struct ndis_tcp_lso_info *lso_info;
 366         int  hdr_offset;
 367         u32 net_trans_info;
 368         u32 hash;
 369         u32 skb_length;
 370         struct hv_page_buffer page_buf[MAX_PAGE_BUFFER_COUNT];
 371         struct hv_page_buffer *pb = page_buf;
 372         struct netvsc_stats *tx_stats = this_cpu_ptr(net_device_ctx->tx_stats);
 373
 374         /* We will atmost need two pages to describe the rndis
 375          * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
 376          * of pages in a single packet. If skb is scattered around
 377          * more pages we try linearizing it.
 378          */
 379
 380 check_size:
 381         skb_length = skb->len;
 382         num_data_pgs = netvsc_get_slots(skb) + 2;
 383         if (num_data_pgs > MAX_PAGE_BUFFER_COUNT && linear) {
 384                 net_alert_ratelimited("packet too big: %u pages (%u bytes)\n",
 385                                       num_data_pgs, skb->len);
 386                 ret = -EFAULT;
 387                 goto drop;
 388         } else if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
 389                 if (skb_linearize(skb)) {
 390                         net_alert_ratelimited("failed to linearize skb\n");
 391                         ret = -ENOMEM;
 392                         goto drop;
 393                 }
 394                 linear = true;
 395                 goto check_size;
 396         }
 397
 398         /*
 399          * Place the rndis header in the skb head room and
 400          * the skb->cb will be used for hv_netvsc_packet
 401          * structure.
 402          */
 403         ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
 404         if (ret) {
 405                 netdev_err(net, "unable to alloc hv_netvsc_packet\n");
 406                 ret = -ENOMEM;
 407                 goto drop;
 408         }
 409         /* Use the skb control buffer for building up the packet */
 410         BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
 411                         FIELD_SIZEOF(struct sk_buff, cb));
 412         packet = (struct hv_netvsc_packet *)skb->cb;
 413
 414
 415         packet->q_idx = skb_get_queue_mapping(skb);
 416
 417         packet->total_data_buflen = skb->len;
 418
 419         rndis_msg = (struct rndis_message *)skb->head;
 420
 421         memset(rndis_msg, 0, RNDIS_AND_PPI_SIZE);
 422
 423         isvlan = skb->vlan_tci & VLAN_TAG_PRESENT;
 424
 425         /* Add the rndis header */
 426         rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
 427         rndis_msg->msg_len = packet->total_data_buflen;
 428         rndis_pkt = &rndis_msg->msg.pkt;
 429         rndis_pkt->data_offset = sizeof(struct rndis_packet);
 430         rndis_pkt->data_len = packet->total_data_buflen;
 431         rndis_pkt->per_pkt_info_offset = sizeof(struct rndis_packet);
 432
 433         rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
 434
 435         hash = skb_get_hash_raw(skb);
 436         if (hash != 0 && net->real_num_tx_queues > 1) {
 437                 rndis_msg_size += NDIS_HASH_PPI_SIZE;
 438                 ppi = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
 439                                     NBL_HASH_VALUE);
 440                 *(u32 *)((void *)ppi + ppi->ppi_offset) = hash;
 441         }
 442
 443         if (isvlan) {
 444                 struct ndis_pkt_8021q_info *vlan;
 445
 446                 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
 447                 ppi = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
 448                                         IEEE_8021Q_INFO);
 449                 vlan = (struct ndis_pkt_8021q_info *)((void *)ppi +
 450                                                 ppi->ppi_offset);
 451                 vlan->vlanid = skb->vlan_tci & VLAN_VID_MASK;
 452                 vlan->pri = (skb->vlan_tci & VLAN_PRIO_MASK) >>
 453                                 VLAN_PRIO_SHIFT;
 454         }
 455
 456         net_trans_info = get_net_transport_info(skb, &hdr_offset);
 457         if (net_trans_info == TRANSPORT_INFO_NOT_IP)
 458                 goto do_send;
 459
 460         /*
 461          * Setup the sendside checksum offload only if this is not a
 462          * GSO packet.
 463          */
 464         if (skb_is_gso(skb))
 465                 goto do_lso;
 466
 467         if ((skb->ip_summed == CHECKSUM_NONE) ||
 468             (skb->ip_summed == CHECKSUM_UNNECESSARY))
 469                 goto do_send;
 470
 471         rndis_msg_size += NDIS_CSUM_PPI_SIZE;
 472         ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
 473                             TCPIP_CHKSUM_PKTINFO);
 474
 475         csum_info = (struct ndis_tcp_ip_checksum_info *)((void *)ppi +
 476                         ppi->ppi_offset);
 477
 478         if (net_trans_info & (INFO_IPV4 << 16))
 479                 csum_info->transmit.is_ipv4 = 1;
 480         else
 481                 csum_info->transmit.is_ipv6 = 1;
 482
 483         if (net_trans_info & INFO_TCP) {
 484                 csum_info->transmit.tcp_checksum = 1;
 485                 csum_info->transmit.tcp_header_offset = hdr_offset;
 486         } else if (net_trans_info & INFO_UDP) {
 487                 /* UDP checksum offload is not supported on ws2008r2.
 488                  * Furthermore, on ws2012 and ws2012r2, there are some
 489                  * issues with udp checksum offload from Linux guests.
 490                  * (these are host issues).
 491                  * For now compute the checksum here.
 492                  */
 493                 struct udphdr *uh;
 494                 u16 udp_len;
 495
 496                 ret = skb_cow_head(skb, 0);
 497                 if (ret)
 498                         goto drop;
 499
 500                 uh = udp_hdr(skb);
 501                 udp_len = ntohs(uh->len);
 502                 uh->check = 0;
 503                 uh->check = csum_tcpudp_magic(ip_hdr(skb)->saddr,
 504                                               ip_hdr(skb)->daddr,
 505                                               udp_len, IPPROTO_UDP,
 506                                               csum_partial(uh, udp_len, 0));
 507                 if (uh->check == 0)
 508                         uh->check = CSUM_MANGLED_0;
 509
 510                 csum_info->transmit.udp_checksum = 0;
 511         }
 512         goto do_send;
 513
 514 do_lso:
 515         rndis_msg_size += NDIS_LSO_PPI_SIZE;
 516         ppi = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
 517                             TCP_LARGESEND_PKTINFO);
 518
 519         lso_info = (struct ndis_tcp_lso_info *)((void *)ppi +
 520                         ppi->ppi_offset);
 521
 522         lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
 523         if (net_trans_info & (INFO_IPV4 << 16)) {
 524                 lso_info->lso_v2_transmit.ip_version =
 525                         NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
 526                 ip_hdr(skb)->tot_len = 0;
 527                 ip_hdr(skb)->check = 0;
 528                 tcp_hdr(skb)->check =
 529                 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
 530                                    ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 531         } else {
 532                 lso_info->lso_v2_transmit.ip_version =
 533                         NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
 534                 ipv6_hdr(skb)->payload_len = 0;
 535                 tcp_hdr(skb)->check =
 536                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 537                                 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 538         }
 539         lso_info->lso_v2_transmit.tcp_header_offset = hdr_offset;
 540         lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
 541
 542 do_send:
 543         /* Start filling in the page buffers with the rndis hdr */
 544         rndis_msg->msg_len += rndis_msg_size;
 545         packet->total_data_buflen = rndis_msg->msg_len;
 546         packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
 547                                                skb, packet, &pb);
 548
 549         /* timestamp packet in software */
 550         skb_tx_timestamp(skb);
 551         ret = netvsc_send(net_device_ctx->device_ctx, packet,
 552                           rndis_msg, &pb, skb);
 553
 554 drop:
 555         if (ret == 0) {
 556                 u64_stats_update_begin(&tx_stats->syncp);
 557                 tx_stats->packets++;
 558                 tx_stats->bytes += skb_length;
 559                 u64_stats_update_end(&tx_stats->syncp);
 560         } else {
 561                 if (ret != -EAGAIN) {
 562                         dev_kfree_skb_any(skb);
 563                         net->stats.tx_dropped++;
 564                 }
 565         }
 566
 567         return (ret == -EAGAIN) ? NETDEV_TX_BUSY : NETDEV_TX_OK;
 568 }
 569
 570 /*
 571  * netvsc_linkstatus_callback - Link up/down notification
 572  */
 573 void netvsc_linkstatus_callback(struct hv_device *device_obj,
 574                                 struct rndis_message *resp)
 575 {
 576         struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
 577         struct net_device *net;
 578         struct net_device_context *ndev_ctx;
 579         struct netvsc_reconfig *event;
 580         unsigned long flags;
 581
 582         /* Handle link change statuses only */
 583         if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
 584             indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
 585             indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
 586                 return;
 587
 588         net = hv_get_drvdata(device_obj);
 589
 590         if (!net || net->reg_state != NETREG_REGISTERED)
 591                 return;
 592
 593         ndev_ctx = netdev_priv(net);
 594
 595         event = kzalloc(sizeof(*event), GFP_ATOMIC);
 596         if (!event)
 597                 return;
 598         event->event = indicate->status;
 599
 600         spin_lock_irqsave(&ndev_ctx->lock, flags);
 601         list_add_tail(&event->list, &ndev_ctx->reconfig_events);
 602         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
 603
 604         schedule_delayed_work(&ndev_ctx->dwork, 0);
 605 }
 606
 607
 608 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
 609                                 struct hv_netvsc_packet *packet,
 610                                 struct ndis_tcp_ip_checksum_info *csum_info,
 611                                 void *data, u16 vlan_tci)
 612 {
 613         struct sk_buff *skb;
 614
 615         skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
 616         if (!skb)
 617                 return skb;
 618
 619         /*
 620          * Copy to skb. This copy is needed here since the memory pointed by
 621          * hv_netvsc_packet cannot be deallocated
 622          */
 623         memcpy(skb_put(skb, packet->total_data_buflen), data,
 624                packet->total_data_buflen);
 625
 626         skb->protocol = eth_type_trans(skb, net);
 627         if (csum_info) {
 628                 /* We only look at the IP checksum here.
 629                  * Should we be dropping the packet if checksum
 630                  * failed? How do we deal with other checksums - TCP/UDP?
 631                  */
 632                 if (csum_info->receive.ip_checksum_succeeded)
 633                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 634                 else
 635                         skb->ip_summed = CHECKSUM_NONE;
 636         }
 637
 638         if (vlan_tci & VLAN_TAG_PRESENT)
 639                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 640                                        vlan_tci);
 641
 642         return skb;
 643 }
 644
 645 /*
 646  * netvsc_recv_callback -  Callback when we receive a packet from the
 647  * "wire" on the specified device.
 648  */
 649 int netvsc_recv_callback(struct hv_device *device_obj,
 650                                 struct hv_netvsc_packet *packet,
 651                                 void **data,
 652                                 struct ndis_tcp_ip_checksum_info *csum_info,
 653                                 struct vmbus_channel *channel,
 654                                 u16 vlan_tci)
 655 {
 656         struct net_device *net = hv_get_drvdata(device_obj);
 657         struct net_device_context *net_device_ctx = netdev_priv(net);
 658         struct sk_buff *skb;
 659         struct sk_buff *vf_skb;
 660         struct netvsc_stats *rx_stats;
 661         struct netvsc_device *netvsc_dev = net_device_ctx->nvdev;
 662         u32 bytes_recvd = packet->total_data_buflen;
 663         int ret = 0;
 664
 665         if (!net || net->reg_state != NETREG_REGISTERED)
 666                 return NVSP_STAT_FAIL;
 667
 668         if (READ_ONCE(netvsc_dev->vf_inject)) {
 669                 atomic_inc(&netvsc_dev->vf_use_cnt);
 670                 if (!READ_ONCE(netvsc_dev->vf_inject)) {
 671                         /*
 672                          * We raced; just move on.
 673                          */
 674                         atomic_dec(&netvsc_dev->vf_use_cnt);
 675                         goto vf_injection_done;
 676                 }
 677
 678                 /*
 679                  * Inject this packet into the VF inerface.
 680                  * On Hyper-V, multicast and brodcast packets
 681                  * are only delivered on the synthetic interface
 682                  * (after subjecting these to policy filters on
 683                  * the host). Deliver these via the VF interface
 684                  * in the guest.
 685                  */
 686                 vf_skb = netvsc_alloc_recv_skb(netvsc_dev->vf_netdev, packet,
 687                                                csum_info, *data, vlan_tci);
 688                 if (vf_skb != NULL) {
 689                         ++netvsc_dev->vf_netdev->stats.rx_packets;
 690                         netvsc_dev->vf_netdev->stats.rx_bytes += bytes_recvd;
 691                         netif_receive_skb(vf_skb);
 692                 } else {
 693                         ++net->stats.rx_dropped;
 694                         ret = NVSP_STAT_FAIL;
 695                 }
 696                 atomic_dec(&netvsc_dev->vf_use_cnt);
 697                 return ret;
 698         }
 699
 700 vf_injection_done:
 701         rx_stats = this_cpu_ptr(net_device_ctx->rx_stats);
 702
 703         /* Allocate a skb - TODO direct I/O to pages? */
 704         skb = netvsc_alloc_recv_skb(net, packet, csum_info, *data, vlan_tci);
 705         if (unlikely(!skb)) {
 706                 ++net->stats.rx_dropped;
 707                 return NVSP_STAT_FAIL;
 708         }
 709         skb_record_rx_queue(skb, channel->
 710                             offermsg.offer.sub_channel_index);
 711
 712         u64_stats_update_begin(&rx_stats->syncp);
 713         rx_stats->packets++;
 714         rx_stats->bytes += packet->total_data_buflen;
 715         u64_stats_update_end(&rx_stats->syncp);
 716
 717         /*
 718          * Pass the skb back up. Network stack will deallocate the skb when it
 719          * is done.
 720          * TODO - use NAPI?
 721          */
 722         netif_rx(skb);
 723
 724         return 0;
 725 }
 726
 727 static void netvsc_get_drvinfo(struct net_device *net,
 728                                struct ethtool_drvinfo *info)
 729 {
 730         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
 731         strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
 732 }
 733
 734 static void netvsc_get_channels(struct net_device *net,
 735                                 struct ethtool_channels *channel)
 736 {
 737         struct net_device_context *net_device_ctx = netdev_priv(net);
 738         struct netvsc_device *nvdev = net_device_ctx->nvdev;
 739
 740         if (nvdev) {
 741                 channel->max_combined   = nvdev->max_chn;
 742                 channel->combined_count = nvdev->num_chn;
 743         }
 744 }
 745
 746 static int netvsc_set_channels(struct net_device *net,
 747                                struct ethtool_channels *channels)
 748 {
 749         struct net_device_context *net_device_ctx = netdev_priv(net);
 750         struct hv_device *dev = net_device_ctx->device_ctx;
 751         struct netvsc_device *nvdev = net_device_ctx->nvdev;
 752         struct netvsc_device_info device_info;
 753         u32 num_chn;
 754         u32 max_chn;
 755         int ret = 0;
 756         bool recovering = false;
 757
 758         if (net_device_ctx->start_remove || !nvdev || nvdev->destroy)
 759                 return -ENODEV;
 760
 761         num_chn = nvdev->num_chn;
 762         max_chn = min_t(u32, nvdev->max_chn, num_online_cpus());
 763
 764         if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5) {
 765                 pr_info("vRSS unsupported before NVSP Version 5\n");
 766                 return -EINVAL;
 767         }
 768
 769         /* We do not support rx, tx, or other */
 770         if (!channels ||
 771             channels->rx_count ||
 772             channels->tx_count ||
 773             channels->other_count ||
 774             (channels->combined_count < 1))
 775                 return -EINVAL;
 776
 777         if (channels->combined_count > max_chn) {
 778                 pr_info("combined channels too high, using %d\n", max_chn);
 779                 channels->combined_count = max_chn;
 780         }
 781
 782         ret = netvsc_close(net);
 783         if (ret)
 784                 goto out;
 785
 786  do_set:
 787         net_device_ctx->start_remove = true;
 788         rndis_filter_device_remove(dev);
 789
 790         nvdev->num_chn = channels->combined_count;
 791
 792         memset(&device_info, 0, sizeof(device_info));
 793         device_info.num_chn = nvdev->num_chn; /* passed to RNDIS */
 794         device_info.ring_size = ring_size;
 795         device_info.max_num_vrss_chns = max_num_vrss_chns;
 796
 797         ret = rndis_filter_device_add(dev, &device_info);
 798         if (ret) {
 799                 if (recovering) {
 800                         netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
 801                         return ret;
 802                 }
 803                 goto recover;
 804         }
 805
 806         nvdev = net_device_ctx->nvdev;
 807
 808         ret = netif_set_real_num_tx_queues(net, nvdev->num_chn);
 809         if (ret) {
 810                 if (recovering) {
 811                         netdev_err(net, "could not set tx queue count (ret %d)\n", ret);
 812                         return ret;
 813                 }
 814                 goto recover;
 815         }
 816
 817         ret = netif_set_real_num_rx_queues(net, nvdev->num_chn);
 818         if (ret) {
 819                 if (recovering) {
 820                         netdev_err(net, "could not set rx queue count (ret %d)\n", ret);
 821                         return ret;
 822                 }
 823                 goto recover;
 824         }
 825
 826  out:
 827         netvsc_open(net);
 828         net_device_ctx->start_remove = false;
 829         /* We may have missed link change notifications */
 830         schedule_delayed_work(&net_device_ctx->dwork, 0);
 831
 832         return ret;
 833
 834  recover:
 835         /* If the above failed, we attempt to recover through the same
 836          * process but with the original number of channels.
 837          */
 838         netdev_err(net, "could not set channels, recovering\n");
 839         recovering = true;
 840         channels->combined_count = num_chn;
 841         goto do_set;
 842 }
 843
 844 static bool netvsc_validate_ethtool_ss_cmd(const struct ethtool_cmd *cmd)
 845 {
 846         struct ethtool_cmd diff1 = *cmd;
 847         struct ethtool_cmd diff2 = {};
 848
 849         ethtool_cmd_speed_set(&diff1, 0);
 850         diff1.duplex = 0;
 851         /* advertising and cmd are usually set */
 852         diff1.advertising = 0;
 853         diff1.cmd = 0;
 854         /* We set port to PORT_OTHER */
 855         diff2.port = PORT_OTHER;
 856
 857         return !memcmp(&diff1, &diff2, sizeof(diff1));
 858 }
 859
 860 static void netvsc_init_settings(struct net_device *dev)
 861 {
 862         struct net_device_context *ndc = netdev_priv(dev);
 863
 864         ndc->speed = SPEED_UNKNOWN;
 865         ndc->duplex = DUPLEX_UNKNOWN;
 866 }
 867
 868 static int netvsc_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
 869 {
 870         struct net_device_context *ndc = netdev_priv(dev);
 871
 872         ethtool_cmd_speed_set(cmd, ndc->speed);
 873         cmd->duplex = ndc->duplex;
 874         cmd->port = PORT_OTHER;
 875
 876         return 0;
 877 }
 878
 879 static int netvsc_set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
 880 {
 881         struct net_device_context *ndc = netdev_priv(dev);
 882         u32 speed;
 883
 884         speed = ethtool_cmd_speed(cmd);
 885         if (!ethtool_validate_speed(speed) ||
 886             !ethtool_validate_duplex(cmd->duplex) ||
 887             !netvsc_validate_ethtool_ss_cmd(cmd))
 888                 return -EINVAL;
 889
 890         ndc->speed = speed;
 891         ndc->duplex = cmd->duplex;
 892
 893         return 0;
 894 }
 895
 896 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
 897 {
 898         struct net_device_context *ndevctx = netdev_priv(ndev);
 899         struct netvsc_device *nvdev = ndevctx->nvdev;
 900         struct hv_device *hdev = ndevctx->device_ctx;
 901         struct netvsc_device_info device_info;
 902         int limit = ETH_DATA_LEN;
 903         u32 num_chn;
 904         int ret = 0;
 905
 906         if (ndevctx->start_remove || !nvdev || nvdev->destroy)
 907                 return -ENODEV;
 908
 909         if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
 910                 limit = NETVSC_MTU - ETH_HLEN;
 911
 912         if (mtu < NETVSC_MTU_MIN || mtu > limit)
 913                 return -EINVAL;
 914
 915         ret = netvsc_close(ndev);
 916         if (ret)
 917                 goto out;
 918
 919         num_chn = nvdev->num_chn;
 920
 921         ndevctx->start_remove = true;
 922         rndis_filter_device_remove(hdev);
 923
 924         ndev->mtu = mtu;
 925
 926         memset(&device_info, 0, sizeof(device_info));
 927         device_info.ring_size = ring_size;
 928         device_info.num_chn = num_chn;
 929         device_info.max_num_vrss_chns = max_num_vrss_chns;
 930         rndis_filter_device_add(hdev, &device_info);
 931
 932 out:
 933         netvsc_open(ndev);
 934         ndevctx->start_remove = false;
 935
 936         /* We may have missed link change notifications */
 937         schedule_delayed_work(&ndevctx->dwork, 0);
 938
 939         return ret;
 940 }
 941
 942 static struct rtnl_link_stats64 *netvsc_get_stats64(struct net_device *net,
 943                                                     struct rtnl_link_stats64 *t)
 944 {
 945         struct net_device_context *ndev_ctx = netdev_priv(net);
 946         int cpu;
 947
 948         for_each_possible_cpu(cpu) {
 949                 struct netvsc_stats *tx_stats = per_cpu_ptr(ndev_ctx->tx_stats,
 950                                                             cpu);
 951                 struct netvsc_stats *rx_stats = per_cpu_ptr(ndev_ctx->rx_stats,
 952                                                             cpu);
 953                 u64 tx_packets, tx_bytes, rx_packets, rx_bytes;
 954                 unsigned int start;
 955
 956                 do {
 957                         start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
 958                         tx_packets = tx_stats->packets;
 959                         tx_bytes = tx_stats->bytes;
 960                 } while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
 961
 962                 do {
 963                         start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
 964                         rx_packets = rx_stats->packets;
 965                         rx_bytes = rx_stats->bytes;
 966                 } while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
 967
 968                 t->tx_bytes     += tx_bytes;
 969                 t->tx_packets   += tx_packets;
 970                 t->rx_bytes     += rx_bytes;
 971                 t->rx_packets   += rx_packets;
 972         }
 973
 974         t->tx_dropped   = net->stats.tx_dropped;
 975         t->tx_errors    = net->stats.tx_dropped;
 976
 977         t->rx_dropped   = net->stats.rx_dropped;
 978         t->rx_errors    = net->stats.rx_errors;
 979
 980         return t;
 981 }
 982
 983 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
 984 {
 985         struct sockaddr *addr = p;
 986         char save_adr[ETH_ALEN];
 987         unsigned char save_aatype;
 988         int err;
 989
 990         memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
 991         save_aatype = ndev->addr_assign_type;
 992
 993         err = eth_mac_addr(ndev, p);
 994         if (err != 0)
 995                 return err;
 996
 997         err = rndis_filter_set_device_mac(ndev, addr->sa_data);
 998         if (err != 0) {
 999                 /* roll back to saved MAC */
1000                 memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
1001                 ndev->addr_assign_type = save_aatype;
1002         }
1003
1004         return err;
1005 }
1006
1007 #ifdef CONFIG_NET_POLL_CONTROLLER
1008 static void netvsc_poll_controller(struct net_device *net)
1009 {
1010         /* As netvsc_start_xmit() works synchronous we don't have to
1011          * trigger anything here.
1012          */
1013 }
1014 #endif
1015
1016 static const struct ethtool_ops ethtool_ops = {
1017         .get_drvinfo    = netvsc_get_drvinfo,
1018         .get_link       = ethtool_op_get_link,
1019         .get_channels   = netvsc_get_channels,
1020         .set_channels   = netvsc_set_channels,
1021         .get_ts_info    = ethtool_op_get_ts_info,
1022         .get_settings   = netvsc_get_settings,
1023         .set_settings   = netvsc_set_settings,
1024 };
1025
1026 static const struct net_device_ops device_ops = {
1027         .ndo_open =                     netvsc_open,
1028         .ndo_stop =                     netvsc_close,
1029         .ndo_start_xmit =               netvsc_start_xmit,
1030         .ndo_set_rx_mode =              netvsc_set_multicast_list,
1031         .ndo_change_mtu =               netvsc_change_mtu,
1032         .ndo_validate_addr =            eth_validate_addr,
1033         .ndo_set_mac_address =          netvsc_set_mac_addr,
1034         .ndo_select_queue =             netvsc_select_queue,
1035         .ndo_get_stats64 =              netvsc_get_stats64,
1036 #ifdef CONFIG_NET_POLL_CONTROLLER
1037         .ndo_poll_controller =          netvsc_poll_controller,
1038 #endif
1039 };
1040
1041 /*
1042  * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
1043  * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
1044  * present send GARP packet to network peers with netif_notify_peers().
1045  */
1046 static void netvsc_link_change(struct work_struct *w)
1047 {
1048         struct net_device_context *ndev_ctx =
1049                 container_of(w, struct net_device_context, dwork.work);
1050         struct hv_device *device_obj = ndev_ctx->device_ctx;
1051         struct net_device *net = hv_get_drvdata(device_obj);
1052         struct netvsc_device *net_device;
1053         struct rndis_device *rdev;
1054         struct netvsc_reconfig *event = NULL;
1055         bool notify = false, reschedule = false;
1056         unsigned long flags, next_reconfig, delay;
1057
1058         rtnl_lock();
1059         if (ndev_ctx->start_remove)
1060                 goto out_unlock;
1061
1062         net_device = ndev_ctx->nvdev;
1063         rdev = net_device->extension;
1064
1065         next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
1066         if (time_is_after_jiffies(next_reconfig)) {
1067                 /* link_watch only sends one notification with current state
1068                  * per second, avoid doing reconfig more frequently. Handle
1069                  * wrap around.
1070                  */
1071                 delay = next_reconfig - jiffies;
1072                 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
1073                 schedule_delayed_work(&ndev_ctx->dwork, delay);
1074                 goto out_unlock;
1075         }
1076         ndev_ctx->last_reconfig = jiffies;
1077
1078         spin_lock_irqsave(&ndev_ctx->lock, flags);
1079         if (!list_empty(&ndev_ctx->reconfig_events)) {
1080                 event = list_first_entry(&ndev_ctx->reconfig_events,
1081                                          struct netvsc_reconfig, list);
1082                 list_del(&event->list);
1083                 reschedule = !list_empty(&ndev_ctx->reconfig_events);
1084         }
1085         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1086
1087         if (!event)
1088                 goto out_unlock;
1089
1090         switch (event->event) {
1091                 /* Only the following events are possible due to the check in
1092                  * netvsc_linkstatus_callback()
1093                  */
1094         case RNDIS_STATUS_MEDIA_CONNECT:
1095                 if (rdev->link_state) {
1096                         rdev->link_state = false;
1097                         netif_carrier_on(net);
1098                         netif_tx_wake_all_queues(net);
1099                 } else {
1100                         notify = true;
1101                 }
1102                 kfree(event);
1103                 break;
1104         case RNDIS_STATUS_MEDIA_DISCONNECT:
1105                 if (!rdev->link_state) {
1106                         rdev->link_state = true;
1107                         netif_carrier_off(net);
1108                         netif_tx_stop_all_queues(net);
1109                 }
1110                 kfree(event);
1111                 break;
1112         case RNDIS_STATUS_NETWORK_CHANGE:
1113                 /* Only makes sense if carrier is present */
1114                 if (!rdev->link_state) {
1115                         rdev->link_state = true;
1116                         netif_carrier_off(net);
1117                         netif_tx_stop_all_queues(net);
1118                         event->event = RNDIS_STATUS_MEDIA_CONNECT;
1119                         spin_lock_irqsave(&ndev_ctx->lock, flags);
1120                         list_add(&event->list, &ndev_ctx->reconfig_events);
1121                         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1122                         reschedule = true;
1123                 }
1124                 break;
1125         }
1126
1127         rtnl_unlock();
1128
1129         if (notify)
1130                 netdev_notify_peers(net);
1131
1132         /* link_watch only sends one notification with current state per
1133          * second, handle next reconfig event in 2 seconds.
1134          */
1135         if (reschedule)
1136                 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1137
1138         return;
1139
1140 out_unlock:
1141         rtnl_unlock();
1142 }
1143
1144 static void netvsc_free_netdev(struct net_device *netdev)
1145 {
1146         struct net_device_context *net_device_ctx = netdev_priv(netdev);
1147
1148         free_percpu(net_device_ctx->tx_stats);
1149         free_percpu(net_device_ctx->rx_stats);
1150         free_netdev(netdev);
1151 }
1152
1153 static void netvsc_notify_peers(struct work_struct *wrk)
1154 {
1155         struct garp_wrk *gwrk;
1156
1157         gwrk = container_of(wrk, struct garp_wrk, dwrk);
1158
1159         netdev_notify_peers(gwrk->netdev);
1160
1161         atomic_dec(&gwrk->netvsc_dev->vf_use_cnt);
1162 }
1163
1164 static struct net_device *get_netvsc_net_device(char *mac)
1165 {
1166         struct net_device *dev, *found = NULL;
1167         int rtnl_locked;
1168
1169         rtnl_locked = rtnl_trylock();
1170
1171         for_each_netdev(&init_net, dev) {
1172                 if (memcmp(dev->dev_addr, mac, ETH_ALEN) == 0) {
1173                         if (dev->netdev_ops != &device_ops)
1174                                 continue;
1175                         found = dev;
1176                         break;
1177                 }
1178         }
1179         if (rtnl_locked)
1180                 rtnl_unlock();
1181
1182         return found;
1183 }
1184
1185 static int netvsc_register_vf(struct net_device *vf_netdev)
1186 {
1187         struct net_device *ndev;
1188         struct net_device_context *net_device_ctx;
1189         struct netvsc_device *netvsc_dev;
1190         const struct ethtool_ops *eth_ops = vf_netdev->ethtool_ops;
1191
1192         if (eth_ops == NULL || eth_ops == &ethtool_ops)
1193                 return NOTIFY_DONE;
1194
1195         /*
1196          * We will use the MAC address to locate the synthetic interface to
1197          * associate with the VF interface. If we don't find a matching
1198          * synthetic interface, move on.
1199          */
1200         ndev = get_netvsc_net_device(vf_netdev->dev_addr);
1201         if (!ndev)
1202                 return NOTIFY_DONE;
1203
1204         net_device_ctx = netdev_priv(ndev);
1205         netvsc_dev = net_device_ctx->nvdev;
1206         if (netvsc_dev == NULL)
1207                 return NOTIFY_DONE;
1208
1209         netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
1210         /*
1211          * Take a reference on the module.
1212          */
1213         try_module_get(THIS_MODULE);
1214         netvsc_dev->vf_netdev = vf_netdev;
1215         return NOTIFY_OK;
1216 }
1217
1218
1219 static int netvsc_vf_up(struct net_device *vf_netdev)
1220 {
1221         struct net_device *ndev;
1222         struct netvsc_device *netvsc_dev;
1223         const struct ethtool_ops *eth_ops = vf_netdev->ethtool_ops;
1224         struct net_device_context *net_device_ctx;
1225
1226         if (eth_ops == &ethtool_ops)
1227                 return NOTIFY_DONE;
1228
1229         ndev = get_netvsc_net_device(vf_netdev->dev_addr);
1230         if (!ndev)
1231                 return NOTIFY_DONE;
1232
1233         net_device_ctx = netdev_priv(ndev);
1234         netvsc_dev = net_device_ctx->nvdev;
1235
1236         if ((netvsc_dev == NULL) || (netvsc_dev->vf_netdev == NULL))
1237                 return NOTIFY_DONE;
1238
1239         netdev_info(ndev, "VF up: %s\n", vf_netdev->name);
1240         netvsc_dev->vf_inject = true;
1241
1242         /*
1243          * Open the device before switching data path.
1244          */
1245         rndis_filter_open(netvsc_dev);
1246
1247         /*
1248          * notify the host to switch the data path.
1249          */
1250         netvsc_switch_datapath(ndev, true);
1251         netdev_info(ndev, "Data path switched to VF: %s\n", vf_netdev->name);
1252
1253         netif_carrier_off(ndev);
1254
1255         /*
1256          * Now notify peers. We are scheduling work to
1257          * notify peers; take a reference to prevent
1258          * the VF interface from vanishing.
1259          */
1260         atomic_inc(&netvsc_dev->vf_use_cnt);
1261         net_device_ctx->gwrk.netdev = vf_netdev;
1262         net_device_ctx->gwrk.netvsc_dev = netvsc_dev;
1263         schedule_work(&net_device_ctx->gwrk.dwrk);
1264
1265         return NOTIFY_OK;
1266 }
1267
1268
1269 static int netvsc_vf_down(struct net_device *vf_netdev)
1270 {
1271         struct net_device *ndev;
1272         struct netvsc_device *netvsc_dev;
1273         struct net_device_context *net_device_ctx;
1274         const struct ethtool_ops *eth_ops = vf_netdev->ethtool_ops;
1275
1276         if (eth_ops == &ethtool_ops)
1277                 return NOTIFY_DONE;
1278
1279         ndev = get_netvsc_net_device(vf_netdev->dev_addr);
1280         if (!ndev)
1281                 return NOTIFY_DONE;
1282
1283         net_device_ctx = netdev_priv(ndev);
1284         netvsc_dev = net_device_ctx->nvdev;
1285
1286         if ((netvsc_dev == NULL) || (netvsc_dev->vf_netdev == NULL))
1287                 return NOTIFY_DONE;
1288
1289         netdev_info(ndev, "VF down: %s\n", vf_netdev->name);
1290         netvsc_dev->vf_inject = false;
1291         /*
1292          * Wait for currently active users to
1293          * drain out.
1294          */
1295
1296         while (atomic_read(&netvsc_dev->vf_use_cnt) != 0)
1297                 udelay(50);
1298         netvsc_switch_datapath(ndev, false);
1299         netdev_info(ndev, "Data path switched from VF: %s\n", vf_netdev->name);
1300         rndis_filter_close(netvsc_dev);
1301         netif_carrier_on(ndev);
1302         /*
1303          * Notify peers.
1304          */
1305         atomic_inc(&netvsc_dev->vf_use_cnt);
1306         net_device_ctx->gwrk.netdev = ndev;
1307         net_device_ctx->gwrk.netvsc_dev = netvsc_dev;
1308         schedule_work(&net_device_ctx->gwrk.dwrk);
1309
1310         return NOTIFY_OK;
1311 }
1312
1313
1314 static int netvsc_unregister_vf(struct net_device *vf_netdev)
1315 {
1316         struct net_device *ndev;
1317         struct netvsc_device *netvsc_dev;
1318         const struct ethtool_ops *eth_ops = vf_netdev->ethtool_ops;
1319         struct net_device_context *net_device_ctx;
1320
1321         if (eth_ops == &ethtool_ops)
1322                 return NOTIFY_DONE;
1323
1324         ndev = get_netvsc_net_device(vf_netdev->dev_addr);
1325         if (!ndev)
1326                 return NOTIFY_DONE;
1327
1328         net_device_ctx = netdev_priv(ndev);
1329         netvsc_dev = net_device_ctx->nvdev;
1330         if (netvsc_dev == NULL)
1331                 return NOTIFY_DONE;
1332         netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
1333
1334         netvsc_dev->vf_netdev = NULL;
1335         module_put(THIS_MODULE);
1336         return NOTIFY_OK;
1337 }
1338
1339 static int netvsc_probe(struct hv_device *dev,
1340                         const struct hv_vmbus_device_id *dev_id)
1341 {
1342         struct net_device *net = NULL;
1343         struct net_device_context *net_device_ctx;
1344         struct netvsc_device_info device_info;
1345         struct netvsc_device *nvdev;
1346         int ret;
1347
1348         net = alloc_etherdev_mq(sizeof(struct net_device_context),
1349                                 num_online_cpus());
1350         if (!net)
1351                 return -ENOMEM;
1352
1353         netif_carrier_off(net);
1354
1355         net_device_ctx = netdev_priv(net);
1356         net_device_ctx->device_ctx = dev;
1357         net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
1358         if (netif_msg_probe(net_device_ctx))
1359                 netdev_dbg(net, "netvsc msg_enable: %d\n",
1360                            net_device_ctx->msg_enable);
1361
1362         net_device_ctx->tx_stats = netdev_alloc_pcpu_stats(struct netvsc_stats);
1363         if (!net_device_ctx->tx_stats) {
1364                 free_netdev(net);
1365                 return -ENOMEM;
1366         }
1367         net_device_ctx->rx_stats = netdev_alloc_pcpu_stats(struct netvsc_stats);
1368         if (!net_device_ctx->rx_stats) {
1369                 free_percpu(net_device_ctx->tx_stats);
1370                 free_netdev(net);
1371                 return -ENOMEM;
1372         }
1373
1374         hv_set_drvdata(dev, net);
1375
1376         net_device_ctx->start_remove = false;
1377
1378         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
1379         INIT_WORK(&net_device_ctx->work, do_set_multicast);
1380         INIT_WORK(&net_device_ctx->gwrk.dwrk, netvsc_notify_peers);
1381
1382         spin_lock_init(&net_device_ctx->lock);
1383         INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
1384
1385         net->netdev_ops = &device_ops;
1386
1387         net->hw_features = NETVSC_HW_FEATURES;
1388         net->features = NETVSC_HW_FEATURES | NETIF_F_HW_VLAN_CTAG_TX;
1389
1390         net->ethtool_ops = &ethtool_ops;
1391         SET_NETDEV_DEV(net, &dev->device);
1392
1393         /* We always need headroom for rndis header */
1394         net->needed_headroom = RNDIS_AND_PPI_SIZE;
1395
1396         /* Notify the netvsc driver of the new device */
1397         memset(&device_info, 0, sizeof(device_info));
1398         device_info.ring_size = ring_size;
1399         device_info.max_num_vrss_chns = max_num_vrss_chns;
1400         ret = rndis_filter_device_add(dev, &device_info);
1401         if (ret != 0) {
1402                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
1403                 netvsc_free_netdev(net);
1404                 hv_set_drvdata(dev, NULL);
1405                 return ret;
1406         }
1407         memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
1408
1409         nvdev = net_device_ctx->nvdev;
1410         netif_set_real_num_tx_queues(net, nvdev->num_chn);
1411         netif_set_real_num_rx_queues(net, nvdev->num_chn);
1412
1413         netvsc_init_settings(net);
1414
1415         ret = register_netdev(net);
1416         if (ret != 0) {
1417                 pr_err("Unable to register netdev.\n");
1418                 rndis_filter_device_remove(dev);
1419                 netvsc_free_netdev(net);
1420         }
1421
1422         return ret;
1423 }
1424
1425 static int netvsc_remove(struct hv_device *dev)
1426 {
1427         struct net_device *net;
1428         struct net_device_context *ndev_ctx;
1429         struct netvsc_device *net_device;
1430
1431         net = hv_get_drvdata(dev);
1432
1433         if (net == NULL) {
1434                 dev_err(&dev->device, "No net device to remove\n");
1435                 return 0;
1436         }
1437
1438
1439         ndev_ctx = netdev_priv(net);
1440         net_device = ndev_ctx->nvdev;
1441
1442         /* Avoid racing with netvsc_change_mtu()/netvsc_set_channels()
1443          * removing the device.
1444          */
1445         rtnl_lock();
1446         ndev_ctx->start_remove = true;
1447         rtnl_unlock();
1448
1449         cancel_delayed_work_sync(&ndev_ctx->dwork);
1450         cancel_work_sync(&ndev_ctx->work);
1451
1452         /* Stop outbound asap */
1453         netif_tx_disable(net);
1454
1455         unregister_netdev(net);
1456
1457         /*
1458          * Call to the vsc driver to let it know that the device is being
1459          * removed
1460          */
1461         rndis_filter_device_remove(dev);
1462
1463         hv_set_drvdata(dev, NULL);
1464
1465         netvsc_free_netdev(net);
1466         return 0;
1467 }
1468
1469 static const struct hv_vmbus_device_id id_table[] = {
1470         /* Network guid */
1471         { HV_NIC_GUID, },
1472         { },
1473 };
1474
1475 MODULE_DEVICE_TABLE(vmbus, id_table);
1476
1477 /* The one and only one */
1478 static struct  hv_driver netvsc_drv = {
1479         .name = KBUILD_MODNAME,
1480         .id_table = id_table,
1481         .probe = netvsc_probe,
1482         .remove = netvsc_remove,
1483 };
1484
1485
1486 /*
1487  * On Hyper-V, every VF interface is matched with a corresponding
1488  * synthetic interface. The synthetic interface is presented first
1489  * to the guest. When the corresponding VF instance is registered,
1490  * we will take care of switching the data path.
1491  */
1492 static int netvsc_netdev_event(struct notifier_block *this,
1493                                unsigned long event, void *ptr)
1494 {
1495         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
1496
1497         /* Avoid Vlan, Bonding dev with same MAC registering as VF */
1498         if (event_dev->priv_flags & (IFF_802_1Q_VLAN | IFF_BONDING))
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);