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 <[email protected]>
  18  *   Hank Janssen  <[email protected]>
  19  */
  20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  21
  22 #include <linux/init.h>
  23 #include <linux/atomic.h>
  24 #include <linux/module.h>
  25 #include <linux/highmem.h>
  26 #include <linux/device.h>
  27 #include <linux/io.h>
  28 #include <linux/delay.h>
  29 #include <linux/netdevice.h>
  30 #include <linux/inetdevice.h>
  31 #include <linux/etherdevice.h>
  32 #include <linux/skbuff.h>
  33 #include <linux/if_vlan.h>
  34 #include <linux/in.h>
  35 #include <linux/slab.h>
  36 #include <net/arp.h>
  37 #include <net/route.h>
  38 #include <net/sock.h>
  39 #include <net/pkt_sched.h>
  40
  41 #include "hyperv_net.h"
  42
  43 #define RING_SIZE_MIN 64
  44 #define LINKCHANGE_INT (2 * HZ)
  45
  46 static int ring_size = 128;
  47 module_param(ring_size, int, S_IRUGO);
  48 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
  49
  50 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
  51                                 NETIF_MSG_LINK | NETIF_MSG_IFUP |
  52                                 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
  53                                 NETIF_MSG_TX_ERR;
  54
  55 static int debug = -1;
  56 module_param(debug, int, S_IRUGO);
  57 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  58
  59 static void do_set_multicast(struct work_struct *w)
  60 {
  61         struct net_device_context *ndevctx =
  62                 container_of(w, struct net_device_context, work);
  63         struct hv_device *device_obj = ndevctx->device_ctx;
  64         struct net_device *ndev = hv_get_drvdata(device_obj);
  65         struct netvsc_device *nvdev = rcu_dereference(ndevctx->nvdev);
  66         struct rndis_device *rdev;
  67
  68         if (!nvdev)
  69                 return;
  70
  71         rdev = nvdev->extension;
  72         if (rdev == NULL)
  73                 return;
  74
  75         if (ndev->flags & IFF_PROMISC)
  76                 rndis_filter_set_packet_filter(rdev,
  77                         NDIS_PACKET_TYPE_PROMISCUOUS);
  78         else
  79                 rndis_filter_set_packet_filter(rdev,
  80                         NDIS_PACKET_TYPE_BROADCAST |
  81                         NDIS_PACKET_TYPE_ALL_MULTICAST |
  82                         NDIS_PACKET_TYPE_DIRECTED);
  83 }
  84
  85 static void netvsc_set_multicast_list(struct net_device *net)
  86 {
  87         struct net_device_context *net_device_ctx = netdev_priv(net);
  88
  89         schedule_work(&net_device_ctx->work);
  90 }
  91
  92 static int netvsc_open(struct net_device *net)
  93 {
  94         struct netvsc_device *nvdev = net_device_to_netvsc_device(net);
  95         struct rndis_device *rdev;
  96         int ret = 0;
  97
  98         netif_carrier_off(net);
  99
 100         /* Open up the device */
 101         ret = rndis_filter_open(nvdev);
 102         if (ret != 0) {
 103                 netdev_err(net, "unable to open device (ret %d).\n", ret);
 104                 return ret;
 105         }
 106
 107         netif_tx_wake_all_queues(net);
 108
 109         rdev = nvdev->extension;
 110         if (!rdev->link_state)
 111                 netif_carrier_on(net);
 112
 113         return ret;
 114 }
 115
 116 static int netvsc_close(struct net_device *net)
 117 {
 118         struct net_device_context *net_device_ctx = netdev_priv(net);
 119         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
 120         int ret;
 121         u32 aread, awrite, i, msec = 10, retry = 0, retry_max = 20;
 122         struct vmbus_channel *chn;
 123
 124         netif_tx_disable(net);
 125
 126         /* Make sure netvsc_set_multicast_list doesn't re-enable filter! */
 127         cancel_work_sync(&net_device_ctx->work);
 128         ret = rndis_filter_close(nvdev);
 129         if (ret != 0) {
 130                 netdev_err(net, "unable to close device (ret %d).\n", ret);
 131                 return ret;
 132         }
 133
 134         /* Ensure pending bytes in ring are read */
 135         while (true) {
 136                 aread = 0;
 137                 for (i = 0; i < nvdev->num_chn; i++) {
 138                         chn = nvdev->chan_table[i].channel;
 139                         if (!chn)
 140                                 continue;
 141
 142                         hv_get_ringbuffer_availbytes(&chn->inbound, &aread,
 143                                                      &awrite);
 144
 145                         if (aread)
 146                                 break;
 147
 148                         hv_get_ringbuffer_availbytes(&chn->outbound, &aread,
 149                                                      &awrite);
 150
 151                         if (aread)
 152                                 break;
 153                 }
 154
 155                 retry++;
 156                 if (retry > retry_max || aread == 0)
 157                         break;
 158
 159                 msleep(msec);
 160
 161                 if (msec < 1000)
 162                         msec *= 2;
 163         }
 164
 165         if (aread) {
 166                 netdev_err(net, "Ring buffer not empty after closing rndis\n");
 167                 ret = -ETIMEDOUT;
 168         }
 169
 170         return ret;
 171 }
 172
 173 static void *init_ppi_data(struct rndis_message *msg, u32 ppi_size,
 174                                 int pkt_type)
 175 {
 176         struct rndis_packet *rndis_pkt;
 177         struct rndis_per_packet_info *ppi;
 178
 179         rndis_pkt = &msg->msg.pkt;
 180         rndis_pkt->data_offset += ppi_size;
 181
 182         ppi = (struct rndis_per_packet_info *)((void *)rndis_pkt +
 183                 rndis_pkt->per_pkt_info_offset + rndis_pkt->per_pkt_info_len);
 184
 185         ppi->size = ppi_size;
 186         ppi->type = pkt_type;
 187         ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
 188
 189         rndis_pkt->per_pkt_info_len += ppi_size;
 190
 191         return ppi;
 192 }
 193
 194 /*
 195  * Select queue for transmit.
 196  *
 197  * If a valid queue has already been assigned, then use that.
 198  * Otherwise compute tx queue based on hash and the send table.
 199  *
 200  * This is basically similar to default (__netdev_pick_tx) with the added step
 201  * of using the host send_table when no other queue has been assigned.
 202  *
 203  * TODO support XPS - but get_xps_queue not exported
 204  */
 205 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
 206                         void *accel_priv, select_queue_fallback_t fallback)
 207 {
 208         struct net_device_context *net_device_ctx = netdev_priv(ndev);
 209         unsigned int num_tx_queues = ndev->real_num_tx_queues;
 210         struct sock *sk = skb->sk;
 211         int q_idx = sk_tx_queue_get(sk);
 212
 213         if (q_idx < 0 || skb->ooo_okay || q_idx >= num_tx_queues) {
 214                 u16 hash = __skb_tx_hash(ndev, skb, VRSS_SEND_TAB_SIZE);
 215                 int new_idx;
 216
 217                 new_idx = net_device_ctx->tx_send_table[hash] % num_tx_queues;
 218
 219                 if (q_idx != new_idx && sk &&
 220                     sk_fullsock(sk) && rcu_access_pointer(sk->sk_dst_cache))
 221                         sk_tx_queue_set(sk, new_idx);
 222
 223                 q_idx = new_idx;
 224         }
 225
 226         return q_idx;
 227 }
 228
 229 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
 230                         struct hv_page_buffer *pb)
 231 {
 232         int j = 0;
 233
 234         /* Deal with compund pages by ignoring unused part
 235          * of the page.
 236          */
 237         page += (offset >> PAGE_SHIFT);
 238         offset &= ~PAGE_MASK;
 239
 240         while (len > 0) {
 241                 unsigned long bytes;
 242
 243                 bytes = PAGE_SIZE - offset;
 244                 if (bytes > len)
 245                         bytes = len;
 246                 pb[j].pfn = page_to_pfn(page);
 247                 pb[j].offset = offset;
 248                 pb[j].len = bytes;
 249
 250                 offset += bytes;
 251                 len -= bytes;
 252
 253                 if (offset == PAGE_SIZE && len) {
 254                         page++;
 255                         offset = 0;
 256                         j++;
 257                 }
 258         }
 259
 260         return j + 1;
 261 }
 262
 263 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
 264                            struct hv_netvsc_packet *packet,
 265                            struct hv_page_buffer **page_buf)
 266 {
 267         struct hv_page_buffer *pb = *page_buf;
 268         u32 slots_used = 0;
 269         char *data = skb->data;
 270         int frags = skb_shinfo(skb)->nr_frags;
 271         int i;
 272
 273         /* The packet is laid out thus:
 274          * 1. hdr: RNDIS header and PPI
 275          * 2. skb linear data
 276          * 3. skb fragment data
 277          */
 278         if (hdr != NULL)
 279                 slots_used += fill_pg_buf(virt_to_page(hdr),
 280                                         offset_in_page(hdr),
 281                                         len, &pb[slots_used]);
 282
 283         packet->rmsg_size = len;
 284         packet->rmsg_pgcnt = slots_used;
 285
 286         slots_used += fill_pg_buf(virt_to_page(data),
 287                                 offset_in_page(data),
 288                                 skb_headlen(skb), &pb[slots_used]);
 289
 290         for (i = 0; i < frags; i++) {
 291                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 292
 293                 slots_used += fill_pg_buf(skb_frag_page(frag),
 294                                         frag->page_offset,
 295                                         skb_frag_size(frag), &pb[slots_used]);
 296         }
 297         return slots_used;
 298 }
 299
 300 static int count_skb_frag_slots(struct sk_buff *skb)
 301 {
 302         int i, frags = skb_shinfo(skb)->nr_frags;
 303         int pages = 0;
 304
 305         for (i = 0; i < frags; i++) {
 306                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 307                 unsigned long size = skb_frag_size(frag);
 308                 unsigned long offset = frag->page_offset;
 309
 310                 /* Skip unused frames from start of page */
 311                 offset &= ~PAGE_MASK;
 312                 pages += PFN_UP(offset + size);
 313         }
 314         return pages;
 315 }
 316
 317 static int netvsc_get_slots(struct sk_buff *skb)
 318 {
 319         char *data = skb->data;
 320         unsigned int offset = offset_in_page(data);
 321         unsigned int len = skb_headlen(skb);
 322         int slots;
 323         int frag_slots;
 324
 325         slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
 326         frag_slots = count_skb_frag_slots(skb);
 327         return slots + frag_slots;
 328 }
 329
 330 static u32 net_checksum_info(struct sk_buff *skb)
 331 {
 332         if (skb->protocol == htons(ETH_P_IP)) {
 333                 struct iphdr *ip = ip_hdr(skb);
 334
 335                 if (ip->protocol == IPPROTO_TCP)
 336                         return TRANSPORT_INFO_IPV4_TCP;
 337                 else if (ip->protocol == IPPROTO_UDP)
 338                         return TRANSPORT_INFO_IPV4_UDP;
 339         } else {
 340                 struct ipv6hdr *ip6 = ipv6_hdr(skb);
 341
 342                 if (ip6->nexthdr == IPPROTO_TCP)
 343                         return TRANSPORT_INFO_IPV6_TCP;
 344                 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
 345                         return TRANSPORT_INFO_IPV6_UDP;
 346         }
 347
 348         return TRANSPORT_INFO_NOT_IP;
 349 }
 350
 351 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
 352 {
 353         struct net_device_context *net_device_ctx = netdev_priv(net);
 354         struct hv_netvsc_packet *packet = NULL;
 355         int ret;
 356         unsigned int num_data_pgs;
 357         struct rndis_message *rndis_msg;
 358         struct rndis_packet *rndis_pkt;
 359         u32 rndis_msg_size;
 360         struct rndis_per_packet_info *ppi;
 361         u32 hash;
 362         struct hv_page_buffer page_buf[MAX_PAGE_BUFFER_COUNT];
 363         struct hv_page_buffer *pb = page_buf;
 364
 365         /* We will atmost need two pages to describe the rndis
 366          * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
 367          * of pages in a single packet. If skb is scattered around
 368          * more pages we try linearizing it.
 369          */
 370
 371         num_data_pgs = netvsc_get_slots(skb) + 2;
 372
 373         if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
 374                 ++net_device_ctx->eth_stats.tx_scattered;
 375
 376                 if (skb_linearize(skb))
 377                         goto no_memory;
 378
 379                 num_data_pgs = netvsc_get_slots(skb) + 2;
 380                 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
 381                         ++net_device_ctx->eth_stats.tx_too_big;
 382                         goto drop;
 383                 }
 384         }
 385
 386         /*
 387          * Place the rndis header in the skb head room and
 388          * the skb->cb will be used for hv_netvsc_packet
 389          * structure.
 390          */
 391         ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
 392         if (ret)
 393                 goto no_memory;
 394
 395         /* Use the skb control buffer for building up the packet */
 396         BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
 397                         FIELD_SIZEOF(struct sk_buff, cb));
 398         packet = (struct hv_netvsc_packet *)skb->cb;
 399
 400         packet->q_idx = skb_get_queue_mapping(skb);
 401
 402         packet->total_data_buflen = skb->len;
 403         packet->total_bytes = skb->len;
 404         packet->total_packets = 1;
 405
 406         rndis_msg = (struct rndis_message *)skb->head;
 407
 408         memset(rndis_msg, 0, RNDIS_AND_PPI_SIZE);
 409
 410         /* Add the rndis header */
 411         rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
 412         rndis_msg->msg_len = packet->total_data_buflen;
 413         rndis_pkt = &rndis_msg->msg.pkt;
 414         rndis_pkt->data_offset = sizeof(struct rndis_packet);
 415         rndis_pkt->data_len = packet->total_data_buflen;
 416         rndis_pkt->per_pkt_info_offset = sizeof(struct rndis_packet);
 417
 418         rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
 419
 420         hash = skb_get_hash_raw(skb);
 421         if (hash != 0 && net->real_num_tx_queues > 1) {
 422                 rndis_msg_size += NDIS_HASH_PPI_SIZE;
 423                 ppi = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
 424                                     NBL_HASH_VALUE);
 425                 *(u32 *)((void *)ppi + ppi->ppi_offset) = hash;
 426         }
 427
 428         if (skb_vlan_tag_present(skb)) {
 429                 struct ndis_pkt_8021q_info *vlan;
 430
 431                 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
 432                 ppi = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
 433                                         IEEE_8021Q_INFO);
 434                 vlan = (struct ndis_pkt_8021q_info *)((void *)ppi +
 435                                                 ppi->ppi_offset);
 436                 vlan->vlanid = skb->vlan_tci & VLAN_VID_MASK;
 437                 vlan->pri = (skb->vlan_tci & VLAN_PRIO_MASK) >>
 438                                 VLAN_PRIO_SHIFT;
 439         }
 440
 441         if (skb_is_gso(skb)) {
 442                 struct ndis_tcp_lso_info *lso_info;
 443
 444                 rndis_msg_size += NDIS_LSO_PPI_SIZE;
 445                 ppi = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
 446                                     TCP_LARGESEND_PKTINFO);
 447
 448                 lso_info = (struct ndis_tcp_lso_info *)((void *)ppi +
 449                                                         ppi->ppi_offset);
 450
 451                 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
 452                 if (skb->protocol == htons(ETH_P_IP)) {
 453                         lso_info->lso_v2_transmit.ip_version =
 454                                 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
 455                         ip_hdr(skb)->tot_len = 0;
 456                         ip_hdr(skb)->check = 0;
 457                         tcp_hdr(skb)->check =
 458                                 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
 459                                                    ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 460                 } else {
 461                         lso_info->lso_v2_transmit.ip_version =
 462                                 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
 463                         ipv6_hdr(skb)->payload_len = 0;
 464                         tcp_hdr(skb)->check =
 465                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 466                                                  &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 467                 }
 468                 lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb);
 469                 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
 470         } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
 471                 if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) {
 472                         struct ndis_tcp_ip_checksum_info *csum_info;
 473
 474                         rndis_msg_size += NDIS_CSUM_PPI_SIZE;
 475                         ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
 476                                             TCPIP_CHKSUM_PKTINFO);
 477
 478                         csum_info = (struct ndis_tcp_ip_checksum_info *)((void *)ppi +
 479                                                                          ppi->ppi_offset);
 480
 481                         csum_info->transmit.tcp_header_offset = skb_transport_offset(skb);
 482
 483                         if (skb->protocol == htons(ETH_P_IP)) {
 484                                 csum_info->transmit.is_ipv4 = 1;
 485
 486                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
 487                                         csum_info->transmit.tcp_checksum = 1;
 488                                 else
 489                                         csum_info->transmit.udp_checksum = 1;
 490                         } else {
 491                                 csum_info->transmit.is_ipv6 = 1;
 492
 493                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
 494                                         csum_info->transmit.tcp_checksum = 1;
 495                                 else
 496                                         csum_info->transmit.udp_checksum = 1;
 497                         }
 498                 } else {
 499                         /* Can't do offload of this type of checksum */
 500                         if (skb_checksum_help(skb))
 501                                 goto drop;
 502                 }
 503         }
 504
 505         /* Start filling in the page buffers with the rndis hdr */
 506         rndis_msg->msg_len += rndis_msg_size;
 507         packet->total_data_buflen = rndis_msg->msg_len;
 508         packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
 509                                                skb, packet, &pb);
 510
 511         /* timestamp packet in software */
 512         skb_tx_timestamp(skb);
 513         ret = netvsc_send(net_device_ctx->device_ctx, packet,
 514                           rndis_msg, &pb, skb);
 515         if (likely(ret == 0))
 516                 return NETDEV_TX_OK;
 517
 518         if (ret == -EAGAIN) {
 519                 ++net_device_ctx->eth_stats.tx_busy;
 520                 return NETDEV_TX_BUSY;
 521         }
 522
 523         if (ret == -ENOSPC)
 524                 ++net_device_ctx->eth_stats.tx_no_space;
 525
 526 drop:
 527         dev_kfree_skb_any(skb);
 528         net->stats.tx_dropped++;
 529
 530         return NETDEV_TX_OK;
 531
 532 no_memory:
 533         ++net_device_ctx->eth_stats.tx_no_memory;
 534         goto drop;
 535 }
 536 /*
 537  * netvsc_linkstatus_callback - Link up/down notification
 538  */
 539 void netvsc_linkstatus_callback(struct hv_device *device_obj,
 540                                 struct rndis_message *resp)
 541 {
 542         struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
 543         struct net_device *net;
 544         struct net_device_context *ndev_ctx;
 545         struct netvsc_reconfig *event;
 546         unsigned long flags;
 547
 548         net = hv_get_drvdata(device_obj);
 549
 550         if (!net)
 551                 return;
 552
 553         ndev_ctx = netdev_priv(net);
 554
 555         /* Update the physical link speed when changing to another vSwitch */
 556         if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
 557                 u32 speed;
 558
 559                 speed = *(u32 *)((void *)indicate + indicate->
 560                                  status_buf_offset) / 10000;
 561                 ndev_ctx->speed = speed;
 562                 return;
 563         }
 564
 565         /* Handle these link change statuses below */
 566         if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
 567             indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
 568             indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
 569                 return;
 570
 571         if (net->reg_state != NETREG_REGISTERED)
 572                 return;
 573
 574         event = kzalloc(sizeof(*event), GFP_ATOMIC);
 575         if (!event)
 576                 return;
 577         event->event = indicate->status;
 578
 579         spin_lock_irqsave(&ndev_ctx->lock, flags);
 580         list_add_tail(&event->list, &ndev_ctx->reconfig_events);
 581         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
 582
 583         schedule_delayed_work(&ndev_ctx->dwork, 0);
 584 }
 585
 586 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
 587                                              struct napi_struct *napi,
 588                                              const struct ndis_tcp_ip_checksum_info *csum_info,
 589                                              const struct ndis_pkt_8021q_info *vlan,
 590                                              void *data, u32 buflen)
 591 {
 592         struct sk_buff *skb;
 593
 594         skb = napi_alloc_skb(napi, buflen);
 595         if (!skb)
 596                 return skb;
 597
 598         /*
 599          * Copy to skb. This copy is needed here since the memory pointed by
 600          * hv_netvsc_packet cannot be deallocated
 601          */
 602         memcpy(skb_put(skb, buflen), data, buflen);
 603
 604         skb->protocol = eth_type_trans(skb, net);
 605
 606         /* skb is already created with CHECKSUM_NONE */
 607         skb_checksum_none_assert(skb);
 608
 609         /*
 610          * In Linux, the IP checksum is always checked.
 611          * Do L4 checksum offload if enabled and present.
 612          */
 613         if (csum_info && (net->features & NETIF_F_RXCSUM)) {
 614                 if (csum_info->receive.tcp_checksum_succeeded ||
 615                     csum_info->receive.udp_checksum_succeeded)
 616                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 617         }
 618
 619         if (vlan) {
 620                 u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT);
 621
 622                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 623                                        vlan_tci);
 624         }
 625
 626         return skb;
 627 }
 628
 629 /*
 630  * netvsc_recv_callback -  Callback when we receive a packet from the
 631  * "wire" on the specified device.
 632  */
 633 int netvsc_recv_callback(struct net_device *net,
 634                          struct vmbus_channel *channel,
 635                          void  *data, u32 len,
 636                          const struct ndis_tcp_ip_checksum_info *csum_info,
 637                          const struct ndis_pkt_8021q_info *vlan)
 638 {
 639         struct net_device_context *net_device_ctx = netdev_priv(net);
 640         struct netvsc_device *net_device;
 641         u16 q_idx = channel->offermsg.offer.sub_channel_index;
 642         struct netvsc_channel *nvchan;
 643         struct net_device *vf_netdev;
 644         struct sk_buff *skb;
 645         struct netvsc_stats *rx_stats;
 646
 647         if (net->reg_state != NETREG_REGISTERED)
 648                 return NVSP_STAT_FAIL;
 649
 650         /*
 651          * If necessary, inject this packet into the VF interface.
 652          * On Hyper-V, multicast and brodcast packets are only delivered
 653          * to the synthetic interface (after subjecting these to
 654          * policy filters on the host). Deliver these via the VF
 655          * interface in the guest.
 656          */
 657         rcu_read_lock();
 658         net_device = rcu_dereference(net_device_ctx->nvdev);
 659         if (unlikely(!net_device))
 660                 goto drop;
 661
 662         nvchan = &net_device->chan_table[q_idx];
 663         vf_netdev = rcu_dereference(net_device_ctx->vf_netdev);
 664         if (vf_netdev && (vf_netdev->flags & IFF_UP))
 665                 net = vf_netdev;
 666
 667         /* Allocate a skb - TODO direct I/O to pages? */
 668         skb = netvsc_alloc_recv_skb(net, &nvchan->napi,
 669                                     csum_info, vlan, data, len);
 670         if (unlikely(!skb)) {
 671 drop:
 672                 ++net->stats.rx_dropped;
 673                 rcu_read_unlock();
 674                 return NVSP_STAT_FAIL;
 675         }
 676
 677         if (net != vf_netdev)
 678                 skb_record_rx_queue(skb, q_idx);
 679
 680         /*
 681          * Even if injecting the packet, record the statistics
 682          * on the synthetic device because modifying the VF device
 683          * statistics will not work correctly.
 684          */
 685         rx_stats = &nvchan->rx_stats;
 686         u64_stats_update_begin(&rx_stats->syncp);
 687         rx_stats->packets++;
 688         rx_stats->bytes += len;
 689
 690         if (skb->pkt_type == PACKET_BROADCAST)
 691                 ++rx_stats->broadcast;
 692         else if (skb->pkt_type == PACKET_MULTICAST)
 693                 ++rx_stats->multicast;
 694         u64_stats_update_end(&rx_stats->syncp);
 695
 696         napi_gro_receive(&nvchan->napi, skb);
 697         rcu_read_unlock();
 698
 699         return 0;
 700 }
 701
 702 static void netvsc_get_drvinfo(struct net_device *net,
 703                                struct ethtool_drvinfo *info)
 704 {
 705         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
 706         strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
 707 }
 708
 709 static void netvsc_get_channels(struct net_device *net,
 710                                 struct ethtool_channels *channel)
 711 {
 712         struct net_device_context *net_device_ctx = netdev_priv(net);
 713         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
 714
 715         if (nvdev) {
 716                 channel->max_combined   = nvdev->max_chn;
 717                 channel->combined_count = nvdev->num_chn;
 718         }
 719 }
 720
 721 static int netvsc_set_queues(struct net_device *net, struct hv_device *dev,
 722                              u32 num_chn)
 723 {
 724         struct netvsc_device_info device_info;
 725         int ret;
 726
 727         memset(&device_info, 0, sizeof(device_info));
 728         device_info.num_chn = num_chn;
 729         device_info.ring_size = ring_size;
 730         device_info.max_num_vrss_chns = num_chn;
 731
 732         ret = rndis_filter_device_add(dev, &device_info);
 733         if (ret)
 734                 return ret;
 735
 736         ret = netif_set_real_num_tx_queues(net, num_chn);
 737         if (ret)
 738                 return ret;
 739
 740         ret = netif_set_real_num_rx_queues(net, num_chn);
 741
 742         return ret;
 743 }
 744
 745 static int netvsc_set_channels(struct net_device *net,
 746                                struct ethtool_channels *channels)
 747 {
 748         struct net_device_context *net_device_ctx = netdev_priv(net);
 749         struct hv_device *dev = net_device_ctx->device_ctx;
 750         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
 751         unsigned int count = channels->combined_count;
 752         bool was_running;
 753         int ret;
 754
 755         /* We do not support separate count for rx, tx, or other */
 756         if (count == 0 ||
 757             channels->rx_count || channels->tx_count || channels->other_count)
 758                 return -EINVAL;
 759
 760         if (count > net->num_tx_queues || count > net->num_rx_queues)
 761                 return -EINVAL;
 762
 763         if (!nvdev || nvdev->destroy)
 764                 return -ENODEV;
 765
 766         if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
 767                 return -EINVAL;
 768
 769         if (count > nvdev->max_chn)
 770                 return -EINVAL;
 771
 772         was_running = netif_running(net);
 773         if (was_running) {
 774                 ret = netvsc_close(net);
 775                 if (ret)
 776                         return ret;
 777         }
 778
 779         rndis_filter_device_remove(dev, nvdev);
 780
 781         ret = netvsc_set_queues(net, dev, count);
 782         if (ret == 0)
 783                 nvdev->num_chn = count;
 784         else
 785                 netvsc_set_queues(net, dev, nvdev->num_chn);
 786
 787         if (was_running)
 788                 ret = netvsc_open(net);
 789
 790         /* We may have missed link change notifications */
 791         schedule_delayed_work(&net_device_ctx->dwork, 0);
 792
 793         return ret;
 794 }
 795
 796 static bool
 797 netvsc_validate_ethtool_ss_cmd(const struct ethtool_link_ksettings *cmd)
 798 {
 799         struct ethtool_link_ksettings diff1 = *cmd;
 800         struct ethtool_link_ksettings diff2 = {};
 801
 802         diff1.base.speed = 0;
 803         diff1.base.duplex = 0;
 804         /* advertising and cmd are usually set */
 805         ethtool_link_ksettings_zero_link_mode(&diff1, advertising);
 806         diff1.base.cmd = 0;
 807         /* We set port to PORT_OTHER */
 808         diff2.base.port = PORT_OTHER;
 809
 810         return !memcmp(&diff1, &diff2, sizeof(diff1));
 811 }
 812
 813 static void netvsc_init_settings(struct net_device *dev)
 814 {
 815         struct net_device_context *ndc = netdev_priv(dev);
 816
 817         ndc->speed = SPEED_UNKNOWN;
 818         ndc->duplex = DUPLEX_UNKNOWN;
 819 }
 820
 821 static int netvsc_get_link_ksettings(struct net_device *dev,
 822                                      struct ethtool_link_ksettings *cmd)
 823 {
 824         struct net_device_context *ndc = netdev_priv(dev);
 825
 826         cmd->base.speed = ndc->speed;
 827         cmd->base.duplex = ndc->duplex;
 828         cmd->base.port = PORT_OTHER;
 829
 830         return 0;
 831 }
 832
 833 static int netvsc_set_link_ksettings(struct net_device *dev,
 834                                      const struct ethtool_link_ksettings *cmd)
 835 {
 836         struct net_device_context *ndc = netdev_priv(dev);
 837         u32 speed;
 838
 839         speed = cmd->base.speed;
 840         if (!ethtool_validate_speed(speed) ||
 841             !ethtool_validate_duplex(cmd->base.duplex) ||
 842             !netvsc_validate_ethtool_ss_cmd(cmd))
 843                 return -EINVAL;
 844
 845         ndc->speed = speed;
 846         ndc->duplex = cmd->base.duplex;
 847
 848         return 0;
 849 }
 850
 851 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
 852 {
 853         struct net_device_context *ndevctx = netdev_priv(ndev);
 854         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
 855         struct hv_device *hdev = ndevctx->device_ctx;
 856         struct netvsc_device_info device_info;
 857         bool was_running;
 858         int ret = 0;
 859
 860         if (!nvdev || nvdev->destroy)
 861                 return -ENODEV;
 862
 863         was_running = netif_running(ndev);
 864         if (was_running) {
 865                 ret = netvsc_close(ndev);
 866                 if (ret)
 867                         return ret;
 868         }
 869
 870         memset(&device_info, 0, sizeof(device_info));
 871         device_info.ring_size = ring_size;
 872         device_info.num_chn = nvdev->num_chn;
 873         device_info.max_num_vrss_chns = nvdev->num_chn;
 874
 875         rndis_filter_device_remove(hdev, nvdev);
 876
 877         /* 'nvdev' has been freed in rndis_filter_device_remove() ->
 878          * netvsc_device_remove () -> free_netvsc_device().
 879          * We mustn't access it before it's re-created in
 880          * rndis_filter_device_add() -> netvsc_device_add().
 881          */
 882
 883         ndev->mtu = mtu;
 884
 885         rndis_filter_device_add(hdev, &device_info);
 886
 887         if (was_running)
 888                 ret = netvsc_open(ndev);
 889
 890         /* We may have missed link change notifications */
 891         schedule_delayed_work(&ndevctx->dwork, 0);
 892
 893         return ret;
 894 }
 895
 896 static void netvsc_get_stats64(struct net_device *net,
 897                                struct rtnl_link_stats64 *t)
 898 {
 899         struct net_device_context *ndev_ctx = netdev_priv(net);
 900         struct netvsc_device *nvdev = rcu_dereference(ndev_ctx->nvdev);
 901         int i;
 902
 903         if (!nvdev)
 904                 return;
 905
 906         for (i = 0; i < nvdev->num_chn; i++) {
 907                 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
 908                 const struct netvsc_stats *stats;
 909                 u64 packets, bytes, multicast;
 910                 unsigned int start;
 911
 912                 stats = &nvchan->tx_stats;
 913                 do {
 914                         start = u64_stats_fetch_begin_irq(&stats->syncp);
 915                         packets = stats->packets;
 916                         bytes = stats->bytes;
 917                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
 918
 919                 t->tx_bytes     += bytes;
 920                 t->tx_packets   += packets;
 921
 922                 stats = &nvchan->rx_stats;
 923                 do {
 924                         start = u64_stats_fetch_begin_irq(&stats->syncp);
 925                         packets = stats->packets;
 926                         bytes = stats->bytes;
 927                         multicast = stats->multicast + stats->broadcast;
 928                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
 929
 930                 t->rx_bytes     += bytes;
 931                 t->rx_packets   += packets;
 932                 t->multicast    += multicast;
 933         }
 934
 935         t->tx_dropped   = net->stats.tx_dropped;
 936         t->tx_errors    = net->stats.tx_errors;
 937
 938         t->rx_dropped   = net->stats.rx_dropped;
 939         t->rx_errors    = net->stats.rx_errors;
 940 }
 941
 942 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
 943 {
 944         struct sockaddr *addr = p;
 945         char save_adr[ETH_ALEN];
 946         unsigned char save_aatype;
 947         int err;
 948
 949         memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
 950         save_aatype = ndev->addr_assign_type;
 951
 952         err = eth_mac_addr(ndev, p);
 953         if (err != 0)
 954                 return err;
 955
 956         err = rndis_filter_set_device_mac(ndev, addr->sa_data);
 957         if (err != 0) {
 958                 /* roll back to saved MAC */
 959                 memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
 960                 ndev->addr_assign_type = save_aatype;
 961         }
 962
 963         return err;
 964 }
 965
 966 static const struct {
 967         char name[ETH_GSTRING_LEN];
 968         u16 offset;
 969 } netvsc_stats[] = {
 970         { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
 971         { "tx_no_memory",  offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
 972         { "tx_no_space",  offsetof(struct netvsc_ethtool_stats, tx_no_space) },
 973         { "tx_too_big",   offsetof(struct netvsc_ethtool_stats, tx_too_big) },
 974         { "tx_busy",      offsetof(struct netvsc_ethtool_stats, tx_busy) },
 975 };
 976
 977 #define NETVSC_GLOBAL_STATS_LEN ARRAY_SIZE(netvsc_stats)
 978
 979 /* 4 statistics per queue (rx/tx packets/bytes) */
 980 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 4)
 981
 982 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
 983 {
 984         struct net_device_context *ndc = netdev_priv(dev);
 985         struct netvsc_device *nvdev = rcu_dereference(ndc->nvdev);
 986
 987         if (!nvdev)
 988                 return -ENODEV;
 989
 990         switch (string_set) {
 991         case ETH_SS_STATS:
 992                 return NETVSC_GLOBAL_STATS_LEN + NETVSC_QUEUE_STATS_LEN(nvdev);
 993         default:
 994                 return -EINVAL;
 995         }
 996 }
 997
 998 static void netvsc_get_ethtool_stats(struct net_device *dev,
 999                                      struct ethtool_stats *stats, u64 *data)
1000 {
1001         struct net_device_context *ndc = netdev_priv(dev);
1002         struct netvsc_device *nvdev = rcu_dereference(ndc->nvdev);
1003         const void *nds = &ndc->eth_stats;
1004         const struct netvsc_stats *qstats;
1005         unsigned int start;
1006         u64 packets, bytes;
1007         int i, j;
1008
1009         if (!nvdev)
1010                 return;
1011
1012         for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
1013                 data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1014
1015         for (j = 0; j < nvdev->num_chn; j++) {
1016                 qstats = &nvdev->chan_table[j].tx_stats;
1017
1018                 do {
1019                         start = u64_stats_fetch_begin_irq(&qstats->syncp);
1020                         packets = qstats->packets;
1021                         bytes = qstats->bytes;
1022                 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1023                 data[i++] = packets;
1024                 data[i++] = bytes;
1025
1026                 qstats = &nvdev->chan_table[j].rx_stats;
1027                 do {
1028                         start = u64_stats_fetch_begin_irq(&qstats->syncp);
1029                         packets = qstats->packets;
1030                         bytes = qstats->bytes;
1031                 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1032                 data[i++] = packets;
1033                 data[i++] = bytes;
1034         }
1035 }
1036
1037 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1038 {
1039         struct net_device_context *ndc = netdev_priv(dev);
1040         struct netvsc_device *nvdev = rcu_dereference(ndc->nvdev);
1041         u8 *p = data;
1042         int i;
1043
1044         if (!nvdev)
1045                 return;
1046
1047         switch (stringset) {
1048         case ETH_SS_STATS:
1049                 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++)
1050                         memcpy(p + i * ETH_GSTRING_LEN,
1051                                netvsc_stats[i].name, ETH_GSTRING_LEN);
1052
1053                 p += i * ETH_GSTRING_LEN;
1054                 for (i = 0; i < nvdev->num_chn; i++) {
1055                         sprintf(p, "tx_queue_%u_packets", i);
1056                         p += ETH_GSTRING_LEN;
1057                         sprintf(p, "tx_queue_%u_bytes", i);
1058                         p += ETH_GSTRING_LEN;
1059                         sprintf(p, "rx_queue_%u_packets", i);
1060                         p += ETH_GSTRING_LEN;
1061                         sprintf(p, "rx_queue_%u_bytes", i);
1062                         p += ETH_GSTRING_LEN;
1063                 }
1064
1065                 break;
1066         }
1067 }
1068
1069 static int
1070 netvsc_get_rss_hash_opts(struct netvsc_device *nvdev,
1071                          struct ethtool_rxnfc *info)
1072 {
1073         info->data = RXH_IP_SRC | RXH_IP_DST;
1074
1075         switch (info->flow_type) {
1076         case TCP_V4_FLOW:
1077         case TCP_V6_FLOW:
1078                 info->data |= RXH_L4_B_0_1 | RXH_L4_B_2_3;
1079                 /* fallthrough */
1080         case UDP_V4_FLOW:
1081         case UDP_V6_FLOW:
1082         case IPV4_FLOW:
1083         case IPV6_FLOW:
1084                 break;
1085         default:
1086                 info->data = 0;
1087                 break;
1088         }
1089
1090         return 0;
1091 }
1092
1093 static int
1094 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1095                  u32 *rules)
1096 {
1097         struct net_device_context *ndc = netdev_priv(dev);
1098         struct netvsc_device *nvdev = rcu_dereference(ndc->nvdev);
1099
1100         if (!nvdev)
1101                 return -ENODEV;
1102
1103         switch (info->cmd) {
1104         case ETHTOOL_GRXRINGS:
1105                 info->data = nvdev->num_chn;
1106                 return 0;
1107
1108         case ETHTOOL_GRXFH:
1109                 return netvsc_get_rss_hash_opts(nvdev, info);
1110         }
1111         return -EOPNOTSUPP;
1112 }
1113
1114 #ifdef CONFIG_NET_POLL_CONTROLLER
1115 static void netvsc_poll_controller(struct net_device *net)
1116 {
1117         /* As netvsc_start_xmit() works synchronous we don't have to
1118          * trigger anything here.
1119          */
1120 }
1121 #endif
1122
1123 static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1124 {
1125         return NETVSC_HASH_KEYLEN;
1126 }
1127
1128 static u32 netvsc_rss_indir_size(struct net_device *dev)
1129 {
1130         return ITAB_NUM;
1131 }
1132
1133 static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key,
1134                            u8 *hfunc)
1135 {
1136         struct net_device_context *ndc = netdev_priv(dev);
1137         struct netvsc_device *ndev = rcu_dereference(ndc->nvdev);
1138         struct rndis_device *rndis_dev;
1139         int i;
1140
1141         if (!ndev)
1142                 return -ENODEV;
1143
1144         if (hfunc)
1145                 *hfunc = ETH_RSS_HASH_TOP;      /* Toeplitz */
1146
1147         rndis_dev = ndev->extension;
1148         if (indir) {
1149                 for (i = 0; i < ITAB_NUM; i++)
1150                         indir[i] = rndis_dev->ind_table[i];
1151         }
1152
1153         if (key)
1154                 memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1155
1156         return 0;
1157 }
1158
1159 static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
1160                            const u8 *key, const u8 hfunc)
1161 {
1162         struct net_device_context *ndc = netdev_priv(dev);
1163         struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1164         struct rndis_device *rndis_dev;
1165         int i;
1166
1167         if (!ndev)
1168                 return -ENODEV;
1169
1170         if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
1171                 return -EOPNOTSUPP;
1172
1173         rndis_dev = ndev->extension;
1174         if (indir) {
1175                 for (i = 0; i < ITAB_NUM; i++)
1176                         if (indir[i] >= dev->num_rx_queues)
1177                                 return -EINVAL;
1178
1179                 for (i = 0; i < ITAB_NUM; i++)
1180                         rndis_dev->ind_table[i] = indir[i];
1181         }
1182
1183         if (!key) {
1184                 if (!indir)
1185                         return 0;
1186
1187                 key = rndis_dev->rss_key;
1188         }
1189
1190         return rndis_filter_set_rss_param(rndis_dev, key, ndev->num_chn);
1191 }
1192
1193 static const struct ethtool_ops ethtool_ops = {
1194         .get_drvinfo    = netvsc_get_drvinfo,
1195         .get_link       = ethtool_op_get_link,
1196         .get_ethtool_stats = netvsc_get_ethtool_stats,
1197         .get_sset_count = netvsc_get_sset_count,
1198         .get_strings    = netvsc_get_strings,
1199         .get_channels   = netvsc_get_channels,
1200         .set_channels   = netvsc_set_channels,
1201         .get_ts_info    = ethtool_op_get_ts_info,
1202         .get_rxnfc      = netvsc_get_rxnfc,
1203         .get_rxfh_key_size = netvsc_get_rxfh_key_size,
1204         .get_rxfh_indir_size = netvsc_rss_indir_size,
1205         .get_rxfh       = netvsc_get_rxfh,
1206         .set_rxfh       = netvsc_set_rxfh,
1207         .get_link_ksettings = netvsc_get_link_ksettings,
1208         .set_link_ksettings = netvsc_set_link_ksettings,
1209 };
1210
1211 static const struct net_device_ops device_ops = {
1212         .ndo_open =                     netvsc_open,
1213         .ndo_stop =                     netvsc_close,
1214         .ndo_start_xmit =               netvsc_start_xmit,
1215         .ndo_set_rx_mode =              netvsc_set_multicast_list,
1216         .ndo_change_mtu =               netvsc_change_mtu,
1217         .ndo_validate_addr =            eth_validate_addr,
1218         .ndo_set_mac_address =          netvsc_set_mac_addr,
1219         .ndo_select_queue =             netvsc_select_queue,
1220         .ndo_get_stats64 =              netvsc_get_stats64,
1221 #ifdef CONFIG_NET_POLL_CONTROLLER
1222         .ndo_poll_controller =          netvsc_poll_controller,
1223 #endif
1224 };
1225
1226 /*
1227  * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
1228  * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
1229  * present send GARP packet to network peers with netif_notify_peers().
1230  */
1231 static void netvsc_link_change(struct work_struct *w)
1232 {
1233         struct net_device_context *ndev_ctx =
1234                 container_of(w, struct net_device_context, dwork.work);
1235         struct hv_device *device_obj = ndev_ctx->device_ctx;
1236         struct net_device *net = hv_get_drvdata(device_obj);
1237         struct netvsc_device *net_device;
1238         struct rndis_device *rdev;
1239         struct netvsc_reconfig *event = NULL;
1240         bool notify = false, reschedule = false;
1241         unsigned long flags, next_reconfig, delay;
1242
1243         rtnl_lock();
1244         net_device = rtnl_dereference(ndev_ctx->nvdev);
1245         if (!net_device)
1246                 goto out_unlock;
1247
1248         rdev = net_device->extension;
1249
1250         next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
1251         if (time_is_after_jiffies(next_reconfig)) {
1252                 /* link_watch only sends one notification with current state
1253                  * per second, avoid doing reconfig more frequently. Handle
1254                  * wrap around.
1255                  */
1256                 delay = next_reconfig - jiffies;
1257                 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
1258                 schedule_delayed_work(&ndev_ctx->dwork, delay);
1259                 goto out_unlock;
1260         }
1261         ndev_ctx->last_reconfig = jiffies;
1262
1263         spin_lock_irqsave(&ndev_ctx->lock, flags);
1264         if (!list_empty(&ndev_ctx->reconfig_events)) {
1265                 event = list_first_entry(&ndev_ctx->reconfig_events,
1266                                          struct netvsc_reconfig, list);
1267                 list_del(&event->list);
1268                 reschedule = !list_empty(&ndev_ctx->reconfig_events);
1269         }
1270         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1271
1272         if (!event)
1273                 goto out_unlock;
1274
1275         switch (event->event) {
1276                 /* Only the following events are possible due to the check in
1277                  * netvsc_linkstatus_callback()
1278                  */
1279         case RNDIS_STATUS_MEDIA_CONNECT:
1280                 if (rdev->link_state) {
1281                         rdev->link_state = false;
1282                         netif_carrier_on(net);
1283                         netif_tx_wake_all_queues(net);
1284                 } else {
1285                         notify = true;
1286                 }
1287                 kfree(event);
1288                 break;
1289         case RNDIS_STATUS_MEDIA_DISCONNECT:
1290                 if (!rdev->link_state) {
1291                         rdev->link_state = true;
1292                         netif_carrier_off(net);
1293                         netif_tx_stop_all_queues(net);
1294                 }
1295                 kfree(event);
1296                 break;
1297         case RNDIS_STATUS_NETWORK_CHANGE:
1298                 /* Only makes sense if carrier is present */
1299                 if (!rdev->link_state) {
1300                         rdev->link_state = true;
1301                         netif_carrier_off(net);
1302                         netif_tx_stop_all_queues(net);
1303                         event->event = RNDIS_STATUS_MEDIA_CONNECT;
1304                         spin_lock_irqsave(&ndev_ctx->lock, flags);
1305                         list_add(&event->list, &ndev_ctx->reconfig_events);
1306                         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1307                         reschedule = true;
1308                 }
1309                 break;
1310         }
1311
1312         rtnl_unlock();
1313
1314         if (notify)
1315                 netdev_notify_peers(net);
1316
1317         /* link_watch only sends one notification with current state per
1318          * second, handle next reconfig event in 2 seconds.
1319          */
1320         if (reschedule)
1321                 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1322
1323         return;
1324
1325 out_unlock:
1326         rtnl_unlock();
1327 }
1328
1329 static struct net_device *get_netvsc_bymac(const u8 *mac)
1330 {
1331         struct net_device *dev;
1332
1333         ASSERT_RTNL();
1334
1335         for_each_netdev(&init_net, dev) {
1336                 if (dev->netdev_ops != &device_ops)
1337                         continue;       /* not a netvsc device */
1338
1339                 if (ether_addr_equal(mac, dev->perm_addr))
1340                         return dev;
1341         }
1342
1343         return NULL;
1344 }
1345
1346 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
1347 {
1348         struct net_device *dev;
1349
1350         ASSERT_RTNL();
1351
1352         for_each_netdev(&init_net, dev) {
1353                 struct net_device_context *net_device_ctx;
1354
1355                 if (dev->netdev_ops != &device_ops)
1356                         continue;       /* not a netvsc device */
1357
1358                 net_device_ctx = netdev_priv(dev);
1359                 if (net_device_ctx->nvdev == NULL)
1360                         continue;       /* device is removed */
1361
1362                 if (rtnl_dereference(net_device_ctx->vf_netdev) == vf_netdev)
1363                         return dev;     /* a match */
1364         }
1365
1366         return NULL;
1367 }
1368
1369 static int netvsc_register_vf(struct net_device *vf_netdev)
1370 {
1371         struct net_device *ndev;
1372         struct net_device_context *net_device_ctx;
1373         struct netvsc_device *netvsc_dev;
1374
1375         if (vf_netdev->addr_len != ETH_ALEN)
1376                 return NOTIFY_DONE;
1377
1378         /*
1379          * We will use the MAC address to locate the synthetic interface to
1380          * associate with the VF interface. If we don't find a matching
1381          * synthetic interface, move on.
1382          */
1383         ndev = get_netvsc_bymac(vf_netdev->perm_addr);
1384         if (!ndev)
1385                 return NOTIFY_DONE;
1386
1387         net_device_ctx = netdev_priv(ndev);
1388         netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
1389         if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
1390                 return NOTIFY_DONE;
1391
1392         netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
1393         /*
1394          * Take a reference on the module.
1395          */
1396         try_module_get(THIS_MODULE);
1397
1398         dev_hold(vf_netdev);
1399         rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
1400         return NOTIFY_OK;
1401 }
1402
1403 static int netvsc_vf_up(struct net_device *vf_netdev)
1404 {
1405         struct net_device *ndev;
1406         struct netvsc_device *netvsc_dev;
1407         struct net_device_context *net_device_ctx;
1408
1409         ndev = get_netvsc_byref(vf_netdev);
1410         if (!ndev)
1411                 return NOTIFY_DONE;
1412
1413         net_device_ctx = netdev_priv(ndev);
1414         netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
1415
1416         netdev_info(ndev, "VF up: %s\n", vf_netdev->name);
1417
1418         /*
1419          * Open the device before switching data path.
1420          */
1421         rndis_filter_open(netvsc_dev);
1422
1423         /*
1424          * notify the host to switch the data path.
1425          */
1426         netvsc_switch_datapath(ndev, true);
1427         netdev_info(ndev, "Data path switched to VF: %s\n", vf_netdev->name);
1428
1429         netif_carrier_off(ndev);
1430
1431         /* Now notify peers through VF device. */
1432         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, vf_netdev);
1433
1434         return NOTIFY_OK;
1435 }
1436
1437 static int netvsc_vf_down(struct net_device *vf_netdev)
1438 {
1439         struct net_device *ndev;
1440         struct netvsc_device *netvsc_dev;
1441         struct net_device_context *net_device_ctx;
1442
1443         ndev = get_netvsc_byref(vf_netdev);
1444         if (!ndev)
1445                 return NOTIFY_DONE;
1446
1447         net_device_ctx = netdev_priv(ndev);
1448         netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
1449
1450         netdev_info(ndev, "VF down: %s\n", vf_netdev->name);
1451         netvsc_switch_datapath(ndev, false);
1452         netdev_info(ndev, "Data path switched from VF: %s\n", vf_netdev->name);
1453         rndis_filter_close(netvsc_dev);
1454         netif_carrier_on(ndev);
1455
1456         /* Now notify peers through netvsc device. */
1457         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, ndev);
1458
1459         return NOTIFY_OK;
1460 }
1461
1462 static int netvsc_unregister_vf(struct net_device *vf_netdev)
1463 {
1464         struct net_device *ndev;
1465         struct net_device_context *net_device_ctx;
1466
1467         ndev = get_netvsc_byref(vf_netdev);
1468         if (!ndev)
1469                 return NOTIFY_DONE;
1470
1471         net_device_ctx = netdev_priv(ndev);
1472
1473         netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
1474
1475         RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
1476         dev_put(vf_netdev);
1477         module_put(THIS_MODULE);
1478         return NOTIFY_OK;
1479 }
1480
1481 static int netvsc_probe(struct hv_device *dev,
1482                         const struct hv_vmbus_device_id *dev_id)
1483 {
1484         struct net_device *net = NULL;
1485         struct net_device_context *net_device_ctx;
1486         struct netvsc_device_info device_info;
1487         struct netvsc_device *nvdev;
1488         int ret;
1489
1490         net = alloc_etherdev_mq(sizeof(struct net_device_context),
1491                                 VRSS_CHANNEL_MAX);
1492         if (!net)
1493                 return -ENOMEM;
1494
1495         netif_carrier_off(net);
1496
1497         netvsc_init_settings(net);
1498
1499         net_device_ctx = netdev_priv(net);
1500         net_device_ctx->device_ctx = dev;
1501         net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
1502         if (netif_msg_probe(net_device_ctx))
1503                 netdev_dbg(net, "netvsc msg_enable: %d\n",
1504                            net_device_ctx->msg_enable);
1505
1506         hv_set_drvdata(dev, net);
1507
1508         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
1509         INIT_WORK(&net_device_ctx->work, do_set_multicast);
1510
1511         spin_lock_init(&net_device_ctx->lock);
1512         INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
1513
1514         net->netdev_ops = &device_ops;
1515         net->ethtool_ops = &ethtool_ops;
1516         SET_NETDEV_DEV(net, &dev->device);
1517
1518         /* We always need headroom for rndis header */
1519         net->needed_headroom = RNDIS_AND_PPI_SIZE;
1520
1521         /* Notify the netvsc driver of the new device */
1522         memset(&device_info, 0, sizeof(device_info));
1523         device_info.ring_size = ring_size;
1524         device_info.num_chn = VRSS_CHANNEL_DEFAULT;
1525         ret = rndis_filter_device_add(dev, &device_info);
1526         if (ret != 0) {
1527                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
1528                 free_netdev(net);
1529                 hv_set_drvdata(dev, NULL);
1530                 return ret;
1531         }
1532         memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
1533
1534         /* hw_features computed in rndis_filter_device_add */
1535         net->features = net->hw_features |
1536                 NETIF_F_HIGHDMA | NETIF_F_SG |
1537                 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
1538         net->vlan_features = net->features;
1539
1540         /* RCU not necessary here, device not registered */
1541         nvdev = net_device_ctx->nvdev;
1542         netif_set_real_num_tx_queues(net, nvdev->num_chn);
1543         netif_set_real_num_rx_queues(net, nvdev->num_chn);
1544
1545         /* MTU range: 68 - 1500 or 65521 */
1546         net->min_mtu = NETVSC_MTU_MIN;
1547         if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
1548                 net->max_mtu = NETVSC_MTU - ETH_HLEN;
1549         else
1550                 net->max_mtu = ETH_DATA_LEN;
1551
1552         ret = register_netdev(net);
1553         if (ret != 0) {
1554                 pr_err("Unable to register netdev.\n");
1555                 rndis_filter_device_remove(dev, nvdev);
1556                 free_netdev(net);
1557         }
1558
1559         return ret;
1560 }
1561
1562 static int netvsc_remove(struct hv_device *dev)
1563 {
1564         struct net_device *net;
1565         struct net_device_context *ndev_ctx;
1566
1567         net = hv_get_drvdata(dev);
1568
1569         if (net == NULL) {
1570                 dev_err(&dev->device, "No net device to remove\n");
1571                 return 0;
1572         }
1573
1574         ndev_ctx = netdev_priv(net);
1575
1576         netif_device_detach(net);
1577
1578         cancel_delayed_work_sync(&ndev_ctx->dwork);
1579         cancel_work_sync(&ndev_ctx->work);
1580
1581         /*
1582          * Call to the vsc driver to let it know that the device is being
1583          * removed. Also blocks mtu and channel changes.
1584          */
1585         rtnl_lock();
1586         rndis_filter_device_remove(dev, ndev_ctx->nvdev);
1587         rtnl_unlock();
1588
1589         unregister_netdev(net);
1590
1591         hv_set_drvdata(dev, NULL);
1592
1593         free_netdev(net);
1594         return 0;
1595 }
1596
1597 static const struct hv_vmbus_device_id id_table[] = {
1598         /* Network guid */
1599         { HV_NIC_GUID, },
1600         { },
1601 };
1602
1603 MODULE_DEVICE_TABLE(vmbus, id_table);
1604
1605 /* The one and only one */
1606 static struct  hv_driver netvsc_drv = {
1607         .name = KBUILD_MODNAME,
1608         .id_table = id_table,
1609         .probe = netvsc_probe,
1610         .remove = netvsc_remove,
1611 };
1612
1613 /*
1614  * On Hyper-V, every VF interface is matched with a corresponding
1615  * synthetic interface. The synthetic interface is presented first
1616  * to the guest. When the corresponding VF instance is registered,
1617  * we will take care of switching the data path.
1618  */
1619 static int netvsc_netdev_event(struct notifier_block *this,
1620                                unsigned long event, void *ptr)
1621 {
1622         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
1623
1624         /* Skip our own events */
1625         if (event_dev->netdev_ops == &device_ops)
1626                 return NOTIFY_DONE;
1627
1628         /* Avoid non-Ethernet type devices */
1629         if (event_dev->type != ARPHRD_ETHER)
1630                 return NOTIFY_DONE;
1631
1632         /* Avoid Vlan dev with same MAC registering as VF */
1633         if (is_vlan_dev(event_dev))
1634                 return NOTIFY_DONE;
1635
1636         /* Avoid Bonding master dev with same MAC registering as VF */
1637         if ((event_dev->priv_flags & IFF_BONDING) &&
1638             (event_dev->flags & IFF_MASTER))
1639                 return NOTIFY_DONE;
1640
1641         switch (event) {
1642         case NETDEV_REGISTER:
1643                 return netvsc_register_vf(event_dev);
1644         case NETDEV_UNREGISTER:
1645                 return netvsc_unregister_vf(event_dev);
1646         case NETDEV_UP:
1647                 return netvsc_vf_up(event_dev);
1648         case NETDEV_DOWN:
1649                 return netvsc_vf_down(event_dev);
1650         default:
1651                 return NOTIFY_DONE;
1652         }
1653 }
1654
1655 static struct notifier_block netvsc_netdev_notifier = {
1656         .notifier_call = netvsc_netdev_event,
1657 };
1658
1659 static void __exit netvsc_drv_exit(void)
1660 {
1661         unregister_netdevice_notifier(&netvsc_netdev_notifier);
1662         vmbus_driver_unregister(&netvsc_drv);
1663 }
1664
1665 static int __init netvsc_drv_init(void)
1666 {
1667         int ret;
1668
1669         if (ring_size < RING_SIZE_MIN) {
1670                 ring_size = RING_SIZE_MIN;
1671                 pr_info("Increased ring_size to %d (min allowed)\n",
1672                         ring_size);
1673         }
1674         ret = vmbus_driver_register(&netvsc_drv);
1675
1676         if (ret)
1677                 return ret;
1678
1679         register_netdevice_notifier(&netvsc_netdev_notifier);
1680         return 0;
1681 }
1682
1683 MODULE_LICENSE("GPL");
1684 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
1685
1686 module_init(netvsc_drv_init);
1687 module_exit(netvsc_drv_exit);