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         net = hv_get_drvdata(device_obj);
 583
 584         if (!net)
 585                 return;
 586
 587         ndev_ctx = netdev_priv(net);
 588
 589         /* Update the physical link speed when changing to another vSwitch */
 590         if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
 591                 u32 speed;
 592
 593                 speed = *(u32 *)((void *)indicate + indicate->
 594                                  status_buf_offset) / 10000;
 595                 ndev_ctx->speed = speed;
 596                 return;
 597         }
 598
 599         /* Handle these link change statuses below */
 600         if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
 601             indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
 602             indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
 603                 return;
 604
 605         if (net->reg_state != NETREG_REGISTERED)
 606                 return;
 607
 608         event = kzalloc(sizeof(*event), GFP_ATOMIC);
 609         if (!event)
 610                 return;
 611         event->event = indicate->status;
 612
 613         spin_lock_irqsave(&ndev_ctx->lock, flags);
 614         list_add_tail(&event->list, &ndev_ctx->reconfig_events);
 615         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
 616
 617         schedule_delayed_work(&ndev_ctx->dwork, 0);
 618 }
 619
 620
 621 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
 622                                 struct hv_netvsc_packet *packet,
 623                                 struct ndis_tcp_ip_checksum_info *csum_info,
 624                                 void *data, u16 vlan_tci)
 625 {
 626         struct sk_buff *skb;
 627
 628         skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
 629         if (!skb)
 630                 return skb;
 631
 632         /*
 633          * Copy to skb. This copy is needed here since the memory pointed by
 634          * hv_netvsc_packet cannot be deallocated
 635          */
 636         memcpy(skb_put(skb, packet->total_data_buflen), data,
 637                packet->total_data_buflen);
 638
 639         skb->protocol = eth_type_trans(skb, net);
 640         if (csum_info) {
 641                 /* We only look at the IP checksum here.
 642                  * Should we be dropping the packet if checksum
 643                  * failed? How do we deal with other checksums - TCP/UDP?
 644                  */
 645                 if (csum_info->receive.ip_checksum_succeeded)
 646                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 647                 else
 648                         skb->ip_summed = CHECKSUM_NONE;
 649         }
 650
 651         if (vlan_tci & VLAN_TAG_PRESENT)
 652                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 653                                        vlan_tci);
 654
 655         return skb;
 656 }
 657
 658 /*
 659  * netvsc_recv_callback -  Callback when we receive a packet from the
 660  * "wire" on the specified device.
 661  */
 662 int netvsc_recv_callback(struct hv_device *device_obj,
 663                                 struct hv_netvsc_packet *packet,
 664                                 void **data,
 665                                 struct ndis_tcp_ip_checksum_info *csum_info,
 666                                 struct vmbus_channel *channel,
 667                                 u16 vlan_tci)
 668 {
 669         struct net_device *net = hv_get_drvdata(device_obj);
 670         struct net_device_context *net_device_ctx = netdev_priv(net);
 671         struct sk_buff *skb;
 672         struct sk_buff *vf_skb;
 673         struct netvsc_stats *rx_stats;
 674         u32 bytes_recvd = packet->total_data_buflen;
 675         int ret = 0;
 676
 677         if (!net || net->reg_state != NETREG_REGISTERED)
 678                 return NVSP_STAT_FAIL;
 679
 680         if (READ_ONCE(net_device_ctx->vf_inject)) {
 681                 atomic_inc(&net_device_ctx->vf_use_cnt);
 682                 if (!READ_ONCE(net_device_ctx->vf_inject)) {
 683                         /*
 684                          * We raced; just move on.
 685                          */
 686                         atomic_dec(&net_device_ctx->vf_use_cnt);
 687                         goto vf_injection_done;
 688                 }
 689
 690                 /*
 691                  * Inject this packet into the VF inerface.
 692                  * On Hyper-V, multicast and brodcast packets
 693                  * are only delivered on the synthetic interface
 694                  * (after subjecting these to policy filters on
 695                  * the host). Deliver these via the VF interface
 696                  * in the guest.
 697                  */
 698                 vf_skb = netvsc_alloc_recv_skb(net_device_ctx->vf_netdev,
 699                                                packet, csum_info, *data,
 700                                                vlan_tci);
 701                 if (vf_skb != NULL) {
 702                         ++net_device_ctx->vf_netdev->stats.rx_packets;
 703                         net_device_ctx->vf_netdev->stats.rx_bytes +=
 704                                 bytes_recvd;
 705                         netif_receive_skb(vf_skb);
 706                 } else {
 707                         ++net->stats.rx_dropped;
 708                         ret = NVSP_STAT_FAIL;
 709                 }
 710                 atomic_dec(&net_device_ctx->vf_use_cnt);
 711                 return ret;
 712         }
 713
 714 vf_injection_done:
 715         rx_stats = this_cpu_ptr(net_device_ctx->rx_stats);
 716
 717         /* Allocate a skb - TODO direct I/O to pages? */
 718         skb = netvsc_alloc_recv_skb(net, packet, csum_info, *data, vlan_tci);
 719         if (unlikely(!skb)) {
 720                 ++net->stats.rx_dropped;
 721                 return NVSP_STAT_FAIL;
 722         }
 723         skb_record_rx_queue(skb, channel->
 724                             offermsg.offer.sub_channel_index);
 725
 726         u64_stats_update_begin(&rx_stats->syncp);
 727         rx_stats->packets++;
 728         rx_stats->bytes += packet->total_data_buflen;
 729         u64_stats_update_end(&rx_stats->syncp);
 730
 731         /*
 732          * Pass the skb back up. Network stack will deallocate the skb when it
 733          * is done.
 734          * TODO - use NAPI?
 735          */
 736         netif_rx(skb);
 737
 738         return 0;
 739 }
 740
 741 static void netvsc_get_drvinfo(struct net_device *net,
 742                                struct ethtool_drvinfo *info)
 743 {
 744         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
 745         strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
 746 }
 747
 748 static void netvsc_get_channels(struct net_device *net,
 749                                 struct ethtool_channels *channel)
 750 {
 751         struct net_device_context *net_device_ctx = netdev_priv(net);
 752         struct netvsc_device *nvdev = net_device_ctx->nvdev;
 753
 754         if (nvdev) {
 755                 channel->max_combined   = nvdev->max_chn;
 756                 channel->combined_count = nvdev->num_chn;
 757         }
 758 }
 759
 760 static int netvsc_set_channels(struct net_device *net,
 761                                struct ethtool_channels *channels)
 762 {
 763         struct net_device_context *net_device_ctx = netdev_priv(net);
 764         struct hv_device *dev = net_device_ctx->device_ctx;
 765         struct netvsc_device *nvdev = net_device_ctx->nvdev;
 766         struct netvsc_device_info device_info;
 767         u32 num_chn;
 768         u32 max_chn;
 769         int ret = 0;
 770         bool recovering = false;
 771
 772         if (net_device_ctx->start_remove || !nvdev || nvdev->destroy)
 773                 return -ENODEV;
 774
 775         num_chn = nvdev->num_chn;
 776         max_chn = min_t(u32, nvdev->max_chn, num_online_cpus());
 777
 778         if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5) {
 779                 pr_info("vRSS unsupported before NVSP Version 5\n");
 780                 return -EINVAL;
 781         }
 782
 783         /* We do not support rx, tx, or other */
 784         if (!channels ||
 785             channels->rx_count ||
 786             channels->tx_count ||
 787             channels->other_count ||
 788             (channels->combined_count < 1))
 789                 return -EINVAL;
 790
 791         if (channels->combined_count > max_chn) {
 792                 pr_info("combined channels too high, using %d\n", max_chn);
 793                 channels->combined_count = max_chn;
 794         }
 795
 796         ret = netvsc_close(net);
 797         if (ret)
 798                 goto out;
 799
 800  do_set:
 801         net_device_ctx->start_remove = true;
 802         rndis_filter_device_remove(dev);
 803
 804         nvdev->num_chn = channels->combined_count;
 805
 806         memset(&device_info, 0, sizeof(device_info));
 807         device_info.num_chn = nvdev->num_chn; /* passed to RNDIS */
 808         device_info.ring_size = ring_size;
 809         device_info.max_num_vrss_chns = max_num_vrss_chns;
 810
 811         ret = rndis_filter_device_add(dev, &device_info);
 812         if (ret) {
 813                 if (recovering) {
 814                         netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
 815                         return ret;
 816                 }
 817                 goto recover;
 818         }
 819
 820         nvdev = net_device_ctx->nvdev;
 821
 822         ret = netif_set_real_num_tx_queues(net, nvdev->num_chn);
 823         if (ret) {
 824                 if (recovering) {
 825                         netdev_err(net, "could not set tx queue count (ret %d)\n", ret);
 826                         return ret;
 827                 }
 828                 goto recover;
 829         }
 830
 831         ret = netif_set_real_num_rx_queues(net, nvdev->num_chn);
 832         if (ret) {
 833                 if (recovering) {
 834                         netdev_err(net, "could not set rx queue count (ret %d)\n", ret);
 835                         return ret;
 836                 }
 837                 goto recover;
 838         }
 839
 840  out:
 841         netvsc_open(net);
 842         net_device_ctx->start_remove = false;
 843         /* We may have missed link change notifications */
 844         schedule_delayed_work(&net_device_ctx->dwork, 0);
 845
 846         return ret;
 847
 848  recover:
 849         /* If the above failed, we attempt to recover through the same
 850          * process but with the original number of channels.
 851          */
 852         netdev_err(net, "could not set channels, recovering\n");
 853         recovering = true;
 854         channels->combined_count = num_chn;
 855         goto do_set;
 856 }
 857
 858 static bool netvsc_validate_ethtool_ss_cmd(const struct ethtool_cmd *cmd)
 859 {
 860         struct ethtool_cmd diff1 = *cmd;
 861         struct ethtool_cmd diff2 = {};
 862
 863         ethtool_cmd_speed_set(&diff1, 0);
 864         diff1.duplex = 0;
 865         /* advertising and cmd are usually set */
 866         diff1.advertising = 0;
 867         diff1.cmd = 0;
 868         /* We set port to PORT_OTHER */
 869         diff2.port = PORT_OTHER;
 870
 871         return !memcmp(&diff1, &diff2, sizeof(diff1));
 872 }
 873
 874 static void netvsc_init_settings(struct net_device *dev)
 875 {
 876         struct net_device_context *ndc = netdev_priv(dev);
 877
 878         ndc->speed = SPEED_UNKNOWN;
 879         ndc->duplex = DUPLEX_UNKNOWN;
 880 }
 881
 882 static int netvsc_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
 883 {
 884         struct net_device_context *ndc = netdev_priv(dev);
 885
 886         ethtool_cmd_speed_set(cmd, ndc->speed);
 887         cmd->duplex = ndc->duplex;
 888         cmd->port = PORT_OTHER;
 889
 890         return 0;
 891 }
 892
 893 static int netvsc_set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
 894 {
 895         struct net_device_context *ndc = netdev_priv(dev);
 896         u32 speed;
 897
 898         speed = ethtool_cmd_speed(cmd);
 899         if (!ethtool_validate_speed(speed) ||
 900             !ethtool_validate_duplex(cmd->duplex) ||
 901             !netvsc_validate_ethtool_ss_cmd(cmd))
 902                 return -EINVAL;
 903
 904         ndc->speed = speed;
 905         ndc->duplex = cmd->duplex;
 906
 907         return 0;
 908 }
 909
 910 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
 911 {
 912         struct net_device_context *ndevctx = netdev_priv(ndev);
 913         struct netvsc_device *nvdev = ndevctx->nvdev;
 914         struct hv_device *hdev = ndevctx->device_ctx;
 915         struct netvsc_device_info device_info;
 916         int limit = ETH_DATA_LEN;
 917         u32 num_chn;
 918         int ret = 0;
 919
 920         if (ndevctx->start_remove || !nvdev || nvdev->destroy)
 921                 return -ENODEV;
 922
 923         if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
 924                 limit = NETVSC_MTU - ETH_HLEN;
 925
 926         if (mtu < NETVSC_MTU_MIN || mtu > limit)
 927                 return -EINVAL;
 928
 929         ret = netvsc_close(ndev);
 930         if (ret)
 931                 goto out;
 932
 933         num_chn = nvdev->num_chn;
 934
 935         ndevctx->start_remove = true;
 936         rndis_filter_device_remove(hdev);
 937
 938         ndev->mtu = mtu;
 939
 940         memset(&device_info, 0, sizeof(device_info));
 941         device_info.ring_size = ring_size;
 942         device_info.num_chn = num_chn;
 943         device_info.max_num_vrss_chns = max_num_vrss_chns;
 944         rndis_filter_device_add(hdev, &device_info);
 945
 946 out:
 947         netvsc_open(ndev);
 948         ndevctx->start_remove = false;
 949
 950         /* We may have missed link change notifications */
 951         schedule_delayed_work(&ndevctx->dwork, 0);
 952
 953         return ret;
 954 }
 955
 956 static struct rtnl_link_stats64 *netvsc_get_stats64(struct net_device *net,
 957                                                     struct rtnl_link_stats64 *t)
 958 {
 959         struct net_device_context *ndev_ctx = netdev_priv(net);
 960         int cpu;
 961
 962         for_each_possible_cpu(cpu) {
 963                 struct netvsc_stats *tx_stats = per_cpu_ptr(ndev_ctx->tx_stats,
 964                                                             cpu);
 965                 struct netvsc_stats *rx_stats = per_cpu_ptr(ndev_ctx->rx_stats,
 966                                                             cpu);
 967                 u64 tx_packets, tx_bytes, rx_packets, rx_bytes;
 968                 unsigned int start;
 969
 970                 do {
 971                         start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
 972                         tx_packets = tx_stats->packets;
 973                         tx_bytes = tx_stats->bytes;
 974                 } while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
 975
 976                 do {
 977                         start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
 978                         rx_packets = rx_stats->packets;
 979                         rx_bytes = rx_stats->bytes;
 980                 } while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
 981
 982                 t->tx_bytes     += tx_bytes;
 983                 t->tx_packets   += tx_packets;
 984                 t->rx_bytes     += rx_bytes;
 985                 t->rx_packets   += rx_packets;
 986         }
 987
 988         t->tx_dropped   = net->stats.tx_dropped;
 989         t->tx_errors    = net->stats.tx_dropped;
 990
 991         t->rx_dropped   = net->stats.rx_dropped;
 992         t->rx_errors    = net->stats.rx_errors;
 993
 994         return t;
 995 }
 996
 997 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
 998 {
 999         struct sockaddr *addr = p;
1000         char save_adr[ETH_ALEN];
1001         unsigned char save_aatype;
1002         int err;
1003
1004         memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
1005         save_aatype = ndev->addr_assign_type;
1006
1007         err = eth_mac_addr(ndev, p);
1008         if (err != 0)
1009                 return err;
1010
1011         err = rndis_filter_set_device_mac(ndev, addr->sa_data);
1012         if (err != 0) {
1013                 /* roll back to saved MAC */
1014                 memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
1015                 ndev->addr_assign_type = save_aatype;
1016         }
1017
1018         return err;
1019 }
1020
1021 #ifdef CONFIG_NET_POLL_CONTROLLER
1022 static void netvsc_poll_controller(struct net_device *net)
1023 {
1024         /* As netvsc_start_xmit() works synchronous we don't have to
1025          * trigger anything here.
1026          */
1027 }
1028 #endif
1029
1030 static const struct ethtool_ops ethtool_ops = {
1031         .get_drvinfo    = netvsc_get_drvinfo,
1032         .get_link       = ethtool_op_get_link,
1033         .get_channels   = netvsc_get_channels,
1034         .set_channels   = netvsc_set_channels,
1035         .get_ts_info    = ethtool_op_get_ts_info,
1036         .get_settings   = netvsc_get_settings,
1037         .set_settings   = netvsc_set_settings,
1038 };
1039
1040 static const struct net_device_ops device_ops = {
1041         .ndo_open =                     netvsc_open,
1042         .ndo_stop =                     netvsc_close,
1043         .ndo_start_xmit =               netvsc_start_xmit,
1044         .ndo_set_rx_mode =              netvsc_set_multicast_list,
1045         .ndo_change_mtu =               netvsc_change_mtu,
1046         .ndo_validate_addr =            eth_validate_addr,
1047         .ndo_set_mac_address =          netvsc_set_mac_addr,
1048         .ndo_select_queue =             netvsc_select_queue,
1049         .ndo_get_stats64 =              netvsc_get_stats64,
1050 #ifdef CONFIG_NET_POLL_CONTROLLER
1051         .ndo_poll_controller =          netvsc_poll_controller,
1052 #endif
1053 };
1054
1055 /*
1056  * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
1057  * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
1058  * present send GARP packet to network peers with netif_notify_peers().
1059  */
1060 static void netvsc_link_change(struct work_struct *w)
1061 {
1062         struct net_device_context *ndev_ctx =
1063                 container_of(w, struct net_device_context, dwork.work);
1064         struct hv_device *device_obj = ndev_ctx->device_ctx;
1065         struct net_device *net = hv_get_drvdata(device_obj);
1066         struct netvsc_device *net_device;
1067         struct rndis_device *rdev;
1068         struct netvsc_reconfig *event = NULL;
1069         bool notify = false, reschedule = false;
1070         unsigned long flags, next_reconfig, delay;
1071
1072         rtnl_lock();
1073         if (ndev_ctx->start_remove)
1074                 goto out_unlock;
1075
1076         net_device = ndev_ctx->nvdev;
1077         rdev = net_device->extension;
1078
1079         next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
1080         if (time_is_after_jiffies(next_reconfig)) {
1081                 /* link_watch only sends one notification with current state
1082                  * per second, avoid doing reconfig more frequently. Handle
1083                  * wrap around.
1084                  */
1085                 delay = next_reconfig - jiffies;
1086                 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
1087                 schedule_delayed_work(&ndev_ctx->dwork, delay);
1088                 goto out_unlock;
1089         }
1090         ndev_ctx->last_reconfig = jiffies;
1091
1092         spin_lock_irqsave(&ndev_ctx->lock, flags);
1093         if (!list_empty(&ndev_ctx->reconfig_events)) {
1094                 event = list_first_entry(&ndev_ctx->reconfig_events,
1095                                          struct netvsc_reconfig, list);
1096                 list_del(&event->list);
1097                 reschedule = !list_empty(&ndev_ctx->reconfig_events);
1098         }
1099         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1100
1101         if (!event)
1102                 goto out_unlock;
1103
1104         switch (event->event) {
1105                 /* Only the following events are possible due to the check in
1106                  * netvsc_linkstatus_callback()
1107                  */
1108         case RNDIS_STATUS_MEDIA_CONNECT:
1109                 if (rdev->link_state) {
1110                         rdev->link_state = false;
1111                         netif_carrier_on(net);
1112                         netif_tx_wake_all_queues(net);
1113                 } else {
1114                         notify = true;
1115                 }
1116                 kfree(event);
1117                 break;
1118         case RNDIS_STATUS_MEDIA_DISCONNECT:
1119                 if (!rdev->link_state) {
1120                         rdev->link_state = true;
1121                         netif_carrier_off(net);
1122                         netif_tx_stop_all_queues(net);
1123                 }
1124                 kfree(event);
1125                 break;
1126         case RNDIS_STATUS_NETWORK_CHANGE:
1127                 /* Only makes sense if carrier is present */
1128                 if (!rdev->link_state) {
1129                         rdev->link_state = true;
1130                         netif_carrier_off(net);
1131                         netif_tx_stop_all_queues(net);
1132                         event->event = RNDIS_STATUS_MEDIA_CONNECT;
1133                         spin_lock_irqsave(&ndev_ctx->lock, flags);
1134                         list_add(&event->list, &ndev_ctx->reconfig_events);
1135                         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1136                         reschedule = true;
1137                 }
1138                 break;
1139         }
1140
1141         rtnl_unlock();
1142
1143         if (notify)
1144                 netdev_notify_peers(net);
1145
1146         /* link_watch only sends one notification with current state per
1147          * second, handle next reconfig event in 2 seconds.
1148          */
1149         if (reschedule)
1150                 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1151
1152         return;
1153
1154 out_unlock:
1155         rtnl_unlock();
1156 }
1157
1158 static void netvsc_free_netdev(struct net_device *netdev)
1159 {
1160         struct net_device_context *net_device_ctx = netdev_priv(netdev);
1161
1162         free_percpu(net_device_ctx->tx_stats);
1163         free_percpu(net_device_ctx->rx_stats);
1164         free_netdev(netdev);
1165 }
1166
1167 static struct net_device *get_netvsc_net_device(char *mac)
1168 {
1169         struct net_device *dev, *found = NULL;
1170         int rtnl_locked;
1171
1172         rtnl_locked = rtnl_trylock();
1173
1174         for_each_netdev(&init_net, dev) {
1175                 if (memcmp(dev->dev_addr, mac, ETH_ALEN) == 0) {
1176                         if (dev->netdev_ops != &device_ops)
1177                                 continue;
1178                         found = dev;
1179                         break;
1180                 }
1181         }
1182         if (rtnl_locked)
1183                 rtnl_unlock();
1184
1185         return found;
1186 }
1187
1188 static int netvsc_register_vf(struct net_device *vf_netdev)
1189 {
1190         struct net_device *ndev;
1191         struct net_device_context *net_device_ctx;
1192         struct netvsc_device *netvsc_dev;
1193         const struct ethtool_ops *eth_ops = vf_netdev->ethtool_ops;
1194
1195         if (eth_ops == NULL || eth_ops == &ethtool_ops)
1196                 return NOTIFY_DONE;
1197
1198         /*
1199          * We will use the MAC address to locate the synthetic interface to
1200          * associate with the VF interface. If we don't find a matching
1201          * synthetic interface, move on.
1202          */
1203         ndev = get_netvsc_net_device(vf_netdev->dev_addr);
1204         if (!ndev)
1205                 return NOTIFY_DONE;
1206
1207         net_device_ctx = netdev_priv(ndev);
1208         netvsc_dev = net_device_ctx->nvdev;
1209         if (!netvsc_dev || net_device_ctx->vf_netdev)
1210                 return NOTIFY_DONE;
1211
1212         netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
1213         /*
1214          * Take a reference on the module.
1215          */
1216         try_module_get(THIS_MODULE);
1217         net_device_ctx->vf_netdev = vf_netdev;
1218         return NOTIFY_OK;
1219 }
1220
1221 static void netvsc_inject_enable(struct net_device_context *net_device_ctx)
1222 {
1223         net_device_ctx->vf_inject = true;
1224 }
1225
1226 static void netvsc_inject_disable(struct net_device_context *net_device_ctx)
1227 {
1228         net_device_ctx->vf_inject = false;
1229
1230         /* Wait for currently active users to drain out. */
1231         while (atomic_read(&net_device_ctx->vf_use_cnt) != 0)
1232                 udelay(50);
1233 }
1234
1235 static int netvsc_vf_up(struct net_device *vf_netdev)
1236 {
1237         struct net_device *ndev;
1238         struct netvsc_device *netvsc_dev;
1239         const struct ethtool_ops *eth_ops = vf_netdev->ethtool_ops;
1240         struct net_device_context *net_device_ctx;
1241
1242         if (eth_ops == &ethtool_ops)
1243                 return NOTIFY_DONE;
1244
1245         ndev = get_netvsc_net_device(vf_netdev->dev_addr);
1246         if (!ndev)
1247                 return NOTIFY_DONE;
1248
1249         net_device_ctx = netdev_priv(ndev);
1250         netvsc_dev = net_device_ctx->nvdev;
1251
1252         if (!netvsc_dev || !net_device_ctx->vf_netdev)
1253                 return NOTIFY_DONE;
1254
1255         netdev_info(ndev, "VF up: %s\n", vf_netdev->name);
1256         netvsc_inject_enable(net_device_ctx);
1257
1258         /*
1259          * Open the device before switching data path.
1260          */
1261         rndis_filter_open(netvsc_dev);
1262
1263         /*
1264          * notify the host to switch the data path.
1265          */
1266         netvsc_switch_datapath(ndev, true);
1267         netdev_info(ndev, "Data path switched to VF: %s\n", vf_netdev->name);
1268
1269         netif_carrier_off(ndev);
1270
1271         /* Now notify peers through VF device. */
1272         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, vf_netdev);
1273
1274         return NOTIFY_OK;
1275 }
1276
1277
1278 static int netvsc_vf_down(struct net_device *vf_netdev)
1279 {
1280         struct net_device *ndev;
1281         struct netvsc_device *netvsc_dev;
1282         struct net_device_context *net_device_ctx;
1283         const struct ethtool_ops *eth_ops = vf_netdev->ethtool_ops;
1284
1285         if (eth_ops == &ethtool_ops)
1286                 return NOTIFY_DONE;
1287
1288         ndev = get_netvsc_net_device(vf_netdev->dev_addr);
1289         if (!ndev)
1290                 return NOTIFY_DONE;
1291
1292         net_device_ctx = netdev_priv(ndev);
1293         netvsc_dev = net_device_ctx->nvdev;
1294
1295         if (!netvsc_dev || !net_device_ctx->vf_netdev)
1296                 return NOTIFY_DONE;
1297
1298         netdev_info(ndev, "VF down: %s\n", vf_netdev->name);
1299         netvsc_inject_disable(net_device_ctx);
1300         netvsc_switch_datapath(ndev, false);
1301         netdev_info(ndev, "Data path switched from VF: %s\n", vf_netdev->name);
1302         rndis_filter_close(netvsc_dev);
1303         netif_carrier_on(ndev);
1304
1305         /* Now notify peers through netvsc device. */
1306         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, ndev);
1307
1308         return NOTIFY_OK;
1309 }
1310
1311
1312 static int netvsc_unregister_vf(struct net_device *vf_netdev)
1313 {
1314         struct net_device *ndev;
1315         struct netvsc_device *netvsc_dev;
1316         const struct ethtool_ops *eth_ops = vf_netdev->ethtool_ops;
1317         struct net_device_context *net_device_ctx;
1318
1319         if (eth_ops == &ethtool_ops)
1320                 return NOTIFY_DONE;
1321
1322         ndev = get_netvsc_net_device(vf_netdev->dev_addr);
1323         if (!ndev)
1324                 return NOTIFY_DONE;
1325
1326         net_device_ctx = netdev_priv(ndev);
1327         netvsc_dev = net_device_ctx->nvdev;
1328         if (!netvsc_dev || !net_device_ctx->vf_netdev)
1329                 return NOTIFY_DONE;
1330         netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
1331         netvsc_inject_disable(net_device_ctx);
1332         net_device_ctx->vf_netdev = NULL;
1333         module_put(THIS_MODULE);
1334         return NOTIFY_OK;
1335 }
1336
1337 static int netvsc_probe(struct hv_device *dev,
1338                         const struct hv_vmbus_device_id *dev_id)
1339 {
1340         struct net_device *net = NULL;
1341         struct net_device_context *net_device_ctx;
1342         struct netvsc_device_info device_info;
1343         struct netvsc_device *nvdev;
1344         int ret;
1345
1346         net = alloc_etherdev_mq(sizeof(struct net_device_context),
1347                                 num_online_cpus());
1348         if (!net)
1349                 return -ENOMEM;
1350
1351         netif_carrier_off(net);
1352
1353         netvsc_init_settings(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
1381         spin_lock_init(&net_device_ctx->lock);
1382         INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
1383
1384         atomic_set(&net_device_ctx->vf_use_cnt, 0);
1385         net_device_ctx->vf_netdev = NULL;
1386         net_device_ctx->vf_inject = false;
1387
1388         net->netdev_ops = &device_ops;
1389
1390         net->hw_features = NETVSC_HW_FEATURES;
1391         net->features = NETVSC_HW_FEATURES | NETIF_F_HW_VLAN_CTAG_TX;
1392
1393         net->ethtool_ops = &ethtool_ops;
1394         SET_NETDEV_DEV(net, &dev->device);
1395
1396         /* We always need headroom for rndis header */
1397         net->needed_headroom = RNDIS_AND_PPI_SIZE;
1398
1399         /* Notify the netvsc driver of the new device */
1400         memset(&device_info, 0, sizeof(device_info));
1401         device_info.ring_size = ring_size;
1402         device_info.max_num_vrss_chns = max_num_vrss_chns;
1403         ret = rndis_filter_device_add(dev, &device_info);
1404         if (ret != 0) {
1405                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
1406                 netvsc_free_netdev(net);
1407                 hv_set_drvdata(dev, NULL);
1408                 return ret;
1409         }
1410         memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
1411
1412         nvdev = net_device_ctx->nvdev;
1413         netif_set_real_num_tx_queues(net, nvdev->num_chn);
1414         netif_set_real_num_rx_queues(net, nvdev->num_chn);
1415
1416         ret = register_netdev(net);
1417         if (ret != 0) {
1418                 pr_err("Unable to register netdev.\n");
1419                 rndis_filter_device_remove(dev);
1420                 netvsc_free_netdev(net);
1421         }
1422
1423         return ret;
1424 }
1425
1426 static int netvsc_remove(struct hv_device *dev)
1427 {
1428         struct net_device *net;
1429         struct net_device_context *ndev_ctx;
1430         struct netvsc_device *net_device;
1431
1432         net = hv_get_drvdata(dev);
1433
1434         if (net == NULL) {
1435                 dev_err(&dev->device, "No net device to remove\n");
1436                 return 0;
1437         }
1438
1439
1440         ndev_ctx = netdev_priv(net);
1441         net_device = ndev_ctx->nvdev;
1442
1443         /* Avoid racing with netvsc_change_mtu()/netvsc_set_channels()
1444          * removing the device.
1445          */
1446         rtnl_lock();
1447         ndev_ctx->start_remove = true;
1448         rtnl_unlock();
1449
1450         cancel_delayed_work_sync(&ndev_ctx->dwork);
1451         cancel_work_sync(&ndev_ctx->work);
1452
1453         /* Stop outbound asap */
1454         netif_tx_disable(net);
1455
1456         unregister_netdev(net);
1457
1458         /*
1459          * Call to the vsc driver to let it know that the device is being
1460          * removed
1461          */
1462         rndis_filter_device_remove(dev);
1463
1464         hv_set_drvdata(dev, NULL);
1465
1466         netvsc_free_netdev(net);
1467         return 0;
1468 }
1469
1470 static const struct hv_vmbus_device_id id_table[] = {
1471         /* Network guid */
1472         { HV_NIC_GUID, },
1473         { },
1474 };
1475
1476 MODULE_DEVICE_TABLE(vmbus, id_table);
1477
1478 /* The one and only one */
1479 static struct  hv_driver netvsc_drv = {
1480         .name = KBUILD_MODNAME,
1481         .id_table = id_table,
1482         .probe = netvsc_probe,
1483         .remove = netvsc_remove,
1484 };
1485
1486
1487 /*
1488  * On Hyper-V, every VF interface is matched with a corresponding
1489  * synthetic interface. The synthetic interface is presented first
1490  * to the guest. When the corresponding VF instance is registered,
1491  * we will take care of switching the data path.
1492  */
1493 static int netvsc_netdev_event(struct notifier_block *this,
1494                                unsigned long event, void *ptr)
1495 {
1496         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
1497
1498         /* Avoid Vlan dev with same MAC registering as VF */
1499         if (event_dev->priv_flags & IFF_802_1Q_VLAN)
1500                 return NOTIFY_DONE;
1501
1502         /* Avoid Bonding master dev with same MAC registering as VF */
1503         if (event_dev->priv_flags & IFF_BONDING &&
1504             event_dev->flags & IFF_MASTER)
1505                 return NOTIFY_DONE;
1506
1507         switch (event) {
1508         case NETDEV_REGISTER:
1509                 return netvsc_register_vf(event_dev);
1510         case NETDEV_UNREGISTER:
1511                 return netvsc_unregister_vf(event_dev);
1512         case NETDEV_UP:
1513                 return netvsc_vf_up(event_dev);
1514         case NETDEV_DOWN:
1515                 return netvsc_vf_down(event_dev);
1516         default:
1517                 return NOTIFY_DONE;
1518         }
1519 }
1520
1521 static struct notifier_block netvsc_netdev_notifier = {
1522         .notifier_call = netvsc_netdev_event,
1523 };
1524
1525 static void __exit netvsc_drv_exit(void)
1526 {
1527         unregister_netdevice_notifier(&netvsc_netdev_notifier);
1528         vmbus_driver_unregister(&netvsc_drv);
1529 }
1530
1531 static int __init netvsc_drv_init(void)
1532 {
1533         int ret;
1534
1535         if (ring_size < RING_SIZE_MIN) {
1536                 ring_size = RING_SIZE_MIN;
1537                 pr_info("Increased ring_size to %d (min allowed)\n",
1538                         ring_size);
1539         }
1540         ret = vmbus_driver_register(&netvsc_drv);
1541
1542         if (ret)
1543                 return ret;
1544
1545         register_netdevice_notifier(&netvsc_netdev_notifier);
1546         return 0;
1547 }
1548
1549 MODULE_LICENSE("GPL");
1550 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
1551
1552 module_init(netvsc_drv_init);
1553 module_exit(netvsc_drv_exit);