drivers/net/hyperv/netvsc_drv.c

   1 /*
   2  * Copyright (c) 2009, Microsoft Corporation.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms and conditions of the GNU General Public License,
   6  * version 2, as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope it will be useful, but WITHOUT
   9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  11  * more details.
  12  *
  13  * You should have received a copy of the GNU General Public License along with
  14  * this program; if not, see <http://www.gnu.org/licenses/>.
  15  *
  16  * Authors:
  17  *   Haiyang Zhang <haiyangz@microsoft.com>
  18  *   Hank Janssen  <hjanssen@microsoft.com>
  19  */
  20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  21
  22 #include <linux/init.h>
  23 #include <linux/atomic.h>
  24 #include <linux/module.h>
  25 #include <linux/highmem.h>
  26 #include <linux/device.h>
  27 #include <linux/io.h>
  28 #include <linux/delay.h>
  29 #include <linux/netdevice.h>
  30 #include <linux/inetdevice.h>
  31 #include <linux/etherdevice.h>
  32 #include <linux/skbuff.h>
  33 #include <linux/if_vlan.h>
  34 #include <linux/in.h>
  35 #include <linux/slab.h>
  36 #include <net/arp.h>
  37 #include <net/route.h>
  38 #include <net/sock.h>
  39 #include <net/pkt_sched.h>
  40
  41 #include "hyperv_net.h"
  42
  43 #define RING_SIZE_MIN 64
  44 #define LINKCHANGE_INT (2 * HZ)
  45
  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 = 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 = 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                                              const struct ndis_tcp_ip_checksum_info *csum_info,
 588                                              const struct ndis_pkt_8021q_info *vlan,
 589                                              void *data, u32 buflen)
 590 {
 591         struct sk_buff *skb;
 592
 593         skb = netdev_alloc_skb_ip_align(net, buflen);
 594         if (!skb)
 595                 return skb;
 596
 597         /*
 598          * Copy to skb. This copy is needed here since the memory pointed by
 599          * hv_netvsc_packet cannot be deallocated
 600          */
 601         memcpy(skb_put(skb, buflen), data, buflen);
 602
 603         skb->protocol = eth_type_trans(skb, net);
 604
 605         /* skb is already created with CHECKSUM_NONE */
 606         skb_checksum_none_assert(skb);
 607
 608         /*
 609          * In Linux, the IP checksum is always checked.
 610          * Do L4 checksum offload if enabled and present.
 611          */
 612         if (csum_info && (net->features & NETIF_F_RXCSUM)) {
 613                 if (csum_info->receive.tcp_checksum_succeeded ||
 614                     csum_info->receive.udp_checksum_succeeded)
 615                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 616         }
 617
 618         if (vlan) {
 619                 u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT);
 620
 621                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 622                                        vlan_tci);
 623         }
 624
 625         return skb;
 626 }
 627
 628 /*
 629  * netvsc_recv_callback -  Callback when we receive a packet from the
 630  * "wire" on the specified device.
 631  */
 632 int netvsc_recv_callback(struct net_device *net,
 633                          struct vmbus_channel *channel,
 634                          void  *data, u32 len,
 635                          const struct ndis_tcp_ip_checksum_info *csum_info,
 636                          const struct ndis_pkt_8021q_info *vlan)
 637 {
 638         struct net_device_context *net_device_ctx = netdev_priv(net);
 639         struct netvsc_device *net_device = net_device_ctx->nvdev;
 640         struct net_device *vf_netdev;
 641         struct sk_buff *skb;
 642         struct netvsc_stats *rx_stats;
 643         u16 q_idx = channel->offermsg.offer.sub_channel_index;
 644
 645
 646         if (net->reg_state != NETREG_REGISTERED)
 647                 return NVSP_STAT_FAIL;
 648
 649         /*
 650          * If necessary, inject this packet into the VF interface.
 651          * On Hyper-V, multicast and brodcast packets are only delivered
 652          * to the synthetic interface (after subjecting these to
 653          * policy filters on the host). Deliver these via the VF
 654          * interface in the guest.
 655          */
 656         rcu_read_lock();
 657         vf_netdev = rcu_dereference(net_device_ctx->vf_netdev);
 658         if (vf_netdev && (vf_netdev->flags & IFF_UP))
 659                 net = vf_netdev;
 660
 661         /* Allocate a skb - TODO direct I/O to pages? */
 662         skb = netvsc_alloc_recv_skb(net, csum_info, vlan, data, len);
 663         if (unlikely(!skb)) {
 664                 ++net->stats.rx_dropped;
 665                 rcu_read_unlock();
 666                 return NVSP_STAT_FAIL;
 667         }
 668
 669         if (net != vf_netdev)
 670                 skb_record_rx_queue(skb, q_idx);
 671
 672         /*
 673          * Even if injecting the packet, record the statistics
 674          * on the synthetic device because modifying the VF device
 675          * statistics will not work correctly.
 676          */
 677         rx_stats = &net_device->chan_table[q_idx].rx_stats;
 678         u64_stats_update_begin(&rx_stats->syncp);
 679         rx_stats->packets++;
 680         rx_stats->bytes += len;
 681
 682         if (skb->pkt_type == PACKET_BROADCAST)
 683                 ++rx_stats->broadcast;
 684         else if (skb->pkt_type == PACKET_MULTICAST)
 685                 ++rx_stats->multicast;
 686         u64_stats_update_end(&rx_stats->syncp);
 687
 688         /*
 689          * Pass the skb back up. Network stack will deallocate the skb when it
 690          * is done.
 691          * TODO - use NAPI?
 692          */
 693         netif_receive_skb(skb);
 694         rcu_read_unlock();
 695
 696         return 0;
 697 }
 698
 699 static void netvsc_get_drvinfo(struct net_device *net,
 700                                struct ethtool_drvinfo *info)
 701 {
 702         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
 703         strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
 704 }
 705
 706 static void netvsc_get_channels(struct net_device *net,
 707                                 struct ethtool_channels *channel)
 708 {
 709         struct net_device_context *net_device_ctx = netdev_priv(net);
 710         struct netvsc_device *nvdev = net_device_ctx->nvdev;
 711
 712         if (nvdev) {
 713                 channel->max_combined   = nvdev->max_chn;
 714                 channel->combined_count = nvdev->num_chn;
 715         }
 716 }
 717
 718 static int netvsc_set_queues(struct net_device *net, struct hv_device *dev,
 719                              u32 num_chn)
 720 {
 721         struct netvsc_device_info device_info;
 722         int ret;
 723
 724         memset(&device_info, 0, sizeof(device_info));
 725         device_info.num_chn = num_chn;
 726         device_info.ring_size = ring_size;
 727         device_info.max_num_vrss_chns = num_chn;
 728
 729         ret = rndis_filter_device_add(dev, &device_info);
 730         if (ret)
 731                 return ret;
 732
 733         ret = netif_set_real_num_tx_queues(net, num_chn);
 734         if (ret)
 735                 return ret;
 736
 737         ret = netif_set_real_num_rx_queues(net, num_chn);
 738
 739         return ret;
 740 }
 741
 742 static int netvsc_set_channels(struct net_device *net,
 743                                struct ethtool_channels *channels)
 744 {
 745         struct net_device_context *net_device_ctx = netdev_priv(net);
 746         struct hv_device *dev = net_device_ctx->device_ctx;
 747         struct netvsc_device *nvdev = net_device_ctx->nvdev;
 748         unsigned int count = channels->combined_count;
 749         int ret;
 750
 751         /* We do not support separate count for rx, tx, or other */
 752         if (count == 0 ||
 753             channels->rx_count || channels->tx_count || channels->other_count)
 754                 return -EINVAL;
 755
 756         if (count > net->num_tx_queues || count > net->num_rx_queues)
 757                 return -EINVAL;
 758
 759         if (net_device_ctx->start_remove || !nvdev || nvdev->destroy)
 760                 return -ENODEV;
 761
 762         if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
 763                 return -EINVAL;
 764
 765         if (count > nvdev->max_chn)
 766                 return -EINVAL;
 767
 768         ret = netvsc_close(net);
 769         if (ret)
 770                 return ret;
 771
 772         net_device_ctx->start_remove = true;
 773         rndis_filter_device_remove(dev, nvdev);
 774
 775         ret = netvsc_set_queues(net, dev, count);
 776         if (ret == 0)
 777                 nvdev->num_chn = count;
 778         else
 779                 netvsc_set_queues(net, dev, nvdev->num_chn);
 780
 781         netvsc_open(net);
 782         net_device_ctx->start_remove = false;
 783
 784         /* We may have missed link change notifications */
 785         schedule_delayed_work(&net_device_ctx->dwork, 0);
 786
 787         return ret;
 788 }
 789
 790 static bool netvsc_validate_ethtool_ss_cmd(const struct ethtool_cmd *cmd)
 791 {
 792         struct ethtool_cmd diff1 = *cmd;
 793         struct ethtool_cmd diff2 = {};
 794
 795         ethtool_cmd_speed_set(&diff1, 0);
 796         diff1.duplex = 0;
 797         /* advertising and cmd are usually set */
 798         diff1.advertising = 0;
 799         diff1.cmd = 0;
 800         /* We set port to PORT_OTHER */
 801         diff2.port = PORT_OTHER;
 802
 803         return !memcmp(&diff1, &diff2, sizeof(diff1));
 804 }
 805
 806 static void netvsc_init_settings(struct net_device *dev)
 807 {
 808         struct net_device_context *ndc = netdev_priv(dev);
 809
 810         ndc->speed = SPEED_UNKNOWN;
 811         ndc->duplex = DUPLEX_UNKNOWN;
 812 }
 813
 814 static int netvsc_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
 815 {
 816         struct net_device_context *ndc = netdev_priv(dev);
 817
 818         ethtool_cmd_speed_set(cmd, ndc->speed);
 819         cmd->duplex = ndc->duplex;
 820         cmd->port = PORT_OTHER;
 821
 822         return 0;
 823 }
 824
 825 static int netvsc_set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
 826 {
 827         struct net_device_context *ndc = netdev_priv(dev);
 828         u32 speed;
 829
 830         speed = ethtool_cmd_speed(cmd);
 831         if (!ethtool_validate_speed(speed) ||
 832             !ethtool_validate_duplex(cmd->duplex) ||
 833             !netvsc_validate_ethtool_ss_cmd(cmd))
 834                 return -EINVAL;
 835
 836         ndc->speed = speed;
 837         ndc->duplex = cmd->duplex;
 838
 839         return 0;
 840 }
 841
 842 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
 843 {
 844         struct net_device_context *ndevctx = netdev_priv(ndev);
 845         struct netvsc_device *nvdev = ndevctx->nvdev;
 846         struct hv_device *hdev = ndevctx->device_ctx;
 847         struct netvsc_device_info device_info;
 848         int ret;
 849
 850         if (ndevctx->start_remove || !nvdev || nvdev->destroy)
 851                 return -ENODEV;
 852
 853         ret = netvsc_close(ndev);
 854         if (ret)
 855                 goto out;
 856
 857         memset(&device_info, 0, sizeof(device_info));
 858         device_info.ring_size = ring_size;
 859         device_info.num_chn = nvdev->num_chn;
 860         device_info.max_num_vrss_chns = nvdev->num_chn;
 861
 862         ndevctx->start_remove = true;
 863         rndis_filter_device_remove(hdev, nvdev);
 864
 865         /* 'nvdev' has been freed in rndis_filter_device_remove() ->
 866          * netvsc_device_remove () -> free_netvsc_device().
 867          * We mustn't access it before it's re-created in
 868          * rndis_filter_device_add() -> netvsc_device_add().
 869          */
 870
 871         ndev->mtu = mtu;
 872
 873         rndis_filter_device_add(hdev, &device_info);
 874
 875 out:
 876         netvsc_open(ndev);
 877         ndevctx->start_remove = false;
 878
 879         /* We may have missed link change notifications */
 880         schedule_delayed_work(&ndevctx->dwork, 0);
 881
 882         return ret;
 883 }
 884
 885 static void netvsc_get_stats64(struct net_device *net,
 886                                struct rtnl_link_stats64 *t)
 887 {
 888         struct net_device_context *ndev_ctx = netdev_priv(net);
 889         struct netvsc_device *nvdev = ndev_ctx->nvdev;
 890         int i;
 891
 892         if (!nvdev)
 893                 return;
 894
 895         for (i = 0; i < nvdev->num_chn; i++) {
 896                 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
 897                 const struct netvsc_stats *stats;
 898                 u64 packets, bytes, multicast;
 899                 unsigned int start;
 900
 901                 stats = &nvchan->tx_stats;
 902                 do {
 903                         start = u64_stats_fetch_begin_irq(&stats->syncp);
 904                         packets = stats->packets;
 905                         bytes = stats->bytes;
 906                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
 907
 908                 t->tx_bytes     += bytes;
 909                 t->tx_packets   += packets;
 910
 911                 stats = &nvchan->rx_stats;
 912                 do {
 913                         start = u64_stats_fetch_begin_irq(&stats->syncp);
 914                         packets = stats->packets;
 915                         bytes = stats->bytes;
 916                         multicast = stats->multicast + stats->broadcast;
 917                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
 918
 919                 t->rx_bytes     += bytes;
 920                 t->rx_packets   += packets;
 921                 t->multicast    += multicast;
 922         }
 923
 924         t->tx_dropped   = net->stats.tx_dropped;
 925         t->tx_errors    = net->stats.tx_errors;
 926
 927         t->rx_dropped   = net->stats.rx_dropped;
 928         t->rx_errors    = net->stats.rx_errors;
 929 }
 930
 931 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
 932 {
 933         struct sockaddr *addr = p;
 934         char save_adr[ETH_ALEN];
 935         unsigned char save_aatype;
 936         int err;
 937
 938         memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
 939         save_aatype = ndev->addr_assign_type;
 940
 941         err = eth_mac_addr(ndev, p);
 942         if (err != 0)
 943                 return err;
 944
 945         err = rndis_filter_set_device_mac(ndev, addr->sa_data);
 946         if (err != 0) {
 947                 /* roll back to saved MAC */
 948                 memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
 949                 ndev->addr_assign_type = save_aatype;
 950         }
 951
 952         return err;
 953 }
 954
 955 static const struct {
 956         char name[ETH_GSTRING_LEN];
 957         u16 offset;
 958 } netvsc_stats[] = {
 959         { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
 960         { "tx_no_memory",  offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
 961         { "tx_no_space",  offsetof(struct netvsc_ethtool_stats, tx_no_space) },
 962         { "tx_too_big",   offsetof(struct netvsc_ethtool_stats, tx_too_big) },
 963         { "tx_busy",      offsetof(struct netvsc_ethtool_stats, tx_busy) },
 964 };
 965
 966 #define NETVSC_GLOBAL_STATS_LEN ARRAY_SIZE(netvsc_stats)
 967
 968 /* 4 statistics per queue (rx/tx packets/bytes) */
 969 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 4)
 970
 971 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
 972 {
 973         struct net_device_context *ndc = netdev_priv(dev);
 974         struct netvsc_device *nvdev = ndc->nvdev;
 975
 976         switch (string_set) {
 977         case ETH_SS_STATS:
 978                 return NETVSC_GLOBAL_STATS_LEN + NETVSC_QUEUE_STATS_LEN(nvdev);
 979         default:
 980                 return -EINVAL;
 981         }
 982 }
 983
 984 static void netvsc_get_ethtool_stats(struct net_device *dev,
 985                                      struct ethtool_stats *stats, u64 *data)
 986 {
 987         struct net_device_context *ndc = netdev_priv(dev);
 988         struct netvsc_device *nvdev = ndc->nvdev;
 989         const void *nds = &ndc->eth_stats;
 990         const struct netvsc_stats *qstats;
 991         unsigned int start;
 992         u64 packets, bytes;
 993         int i, j;
 994
 995         for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
 996                 data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
 997
 998         for (j = 0; j < nvdev->num_chn; j++) {
 999                 qstats = &nvdev->chan_table[j].tx_stats;
1000
1001                 do {
1002                         start = u64_stats_fetch_begin_irq(&qstats->syncp);
1003                         packets = qstats->packets;
1004                         bytes = qstats->bytes;
1005                 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1006                 data[i++] = packets;
1007                 data[i++] = bytes;
1008
1009                 qstats = &nvdev->chan_table[j].rx_stats;
1010                 do {
1011                         start = u64_stats_fetch_begin_irq(&qstats->syncp);
1012                         packets = qstats->packets;
1013                         bytes = qstats->bytes;
1014                 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1015                 data[i++] = packets;
1016                 data[i++] = bytes;
1017         }
1018 }
1019
1020 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1021 {
1022         struct net_device_context *ndc = netdev_priv(dev);
1023         struct netvsc_device *nvdev = ndc->nvdev;
1024         u8 *p = data;
1025         int i;
1026
1027         switch (stringset) {
1028         case ETH_SS_STATS:
1029                 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++)
1030                         memcpy(p + i * ETH_GSTRING_LEN,
1031                                netvsc_stats[i].name, ETH_GSTRING_LEN);
1032
1033                 p += i * ETH_GSTRING_LEN;
1034                 for (i = 0; i < nvdev->num_chn; i++) {
1035                         sprintf(p, "tx_queue_%u_packets", i);
1036                         p += ETH_GSTRING_LEN;
1037                         sprintf(p, "tx_queue_%u_bytes", i);
1038                         p += ETH_GSTRING_LEN;
1039                         sprintf(p, "rx_queue_%u_packets", i);
1040                         p += ETH_GSTRING_LEN;
1041                         sprintf(p, "rx_queue_%u_bytes", i);
1042                         p += ETH_GSTRING_LEN;
1043                 }
1044
1045                 break;
1046         }
1047 }
1048
1049 static int
1050 netvsc_get_rss_hash_opts(struct netvsc_device *nvdev,
1051                          struct ethtool_rxnfc *info)
1052 {
1053         info->data = RXH_IP_SRC | RXH_IP_DST;
1054
1055         switch (info->flow_type) {
1056         case TCP_V4_FLOW:
1057         case TCP_V6_FLOW:
1058                 info->data |= RXH_L4_B_0_1 | RXH_L4_B_2_3;
1059                 /* fallthrough */
1060         case UDP_V4_FLOW:
1061         case UDP_V6_FLOW:
1062         case IPV4_FLOW:
1063         case IPV6_FLOW:
1064                 break;
1065         default:
1066                 info->data = 0;
1067                 break;
1068         }
1069
1070         return 0;
1071 }
1072
1073 static int
1074 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1075                  u32 *rules)
1076 {
1077         struct net_device_context *ndc = netdev_priv(dev);
1078         struct netvsc_device *nvdev = ndc->nvdev;
1079
1080         switch (info->cmd) {
1081         case ETHTOOL_GRXRINGS:
1082                 info->data = nvdev->num_chn;
1083                 return 0;
1084
1085         case ETHTOOL_GRXFH:
1086                 return netvsc_get_rss_hash_opts(nvdev, info);
1087         }
1088         return -EOPNOTSUPP;
1089 }
1090
1091 #ifdef CONFIG_NET_POLL_CONTROLLER
1092 static void netvsc_poll_controller(struct net_device *net)
1093 {
1094         /* As netvsc_start_xmit() works synchronous we don't have to
1095          * trigger anything here.
1096          */
1097 }
1098 #endif
1099
1100 static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1101 {
1102         return NETVSC_HASH_KEYLEN;
1103 }
1104
1105 static u32 netvsc_rss_indir_size(struct net_device *dev)
1106 {
1107         return ITAB_NUM;
1108 }
1109
1110 static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key,
1111                            u8 *hfunc)
1112 {
1113         struct net_device_context *ndc = netdev_priv(dev);
1114         struct netvsc_device *ndev = ndc->nvdev;
1115         struct rndis_device *rndis_dev = ndev->extension;
1116         int i;
1117
1118         if (hfunc)
1119                 *hfunc = ETH_RSS_HASH_TOP;      /* Toeplitz */
1120
1121         if (indir) {
1122                 for (i = 0; i < ITAB_NUM; i++)
1123                         indir[i] = rndis_dev->ind_table[i];
1124         }
1125
1126         if (key)
1127                 memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1128
1129         return 0;
1130 }
1131
1132 static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
1133                            const u8 *key, const u8 hfunc)
1134 {
1135         struct net_device_context *ndc = netdev_priv(dev);
1136         struct netvsc_device *ndev = ndc->nvdev;
1137         struct rndis_device *rndis_dev = ndev->extension;
1138         int i;
1139
1140         if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
1141                 return -EOPNOTSUPP;
1142
1143         if (indir) {
1144                 for (i = 0; i < ITAB_NUM; i++)
1145                         if (indir[i] >= dev->num_rx_queues)
1146                                 return -EINVAL;
1147
1148                 for (i = 0; i < ITAB_NUM; i++)
1149                         rndis_dev->ind_table[i] = indir[i];
1150         }
1151
1152         if (!key) {
1153                 if (!indir)
1154                         return 0;
1155
1156                 key = rndis_dev->rss_key;
1157         }
1158
1159         return rndis_filter_set_rss_param(rndis_dev, key, ndev->num_chn);
1160 }
1161
1162 static const struct ethtool_ops ethtool_ops = {
1163         .get_drvinfo    = netvsc_get_drvinfo,
1164         .get_link       = ethtool_op_get_link,
1165         .get_ethtool_stats = netvsc_get_ethtool_stats,
1166         .get_sset_count = netvsc_get_sset_count,
1167         .get_strings    = netvsc_get_strings,
1168         .get_channels   = netvsc_get_channels,
1169         .set_channels   = netvsc_set_channels,
1170         .get_ts_info    = ethtool_op_get_ts_info,
1171         .get_settings   = netvsc_get_settings,
1172         .set_settings   = netvsc_set_settings,
1173         .get_rxnfc      = netvsc_get_rxnfc,
1174         .get_rxfh_key_size = netvsc_get_rxfh_key_size,
1175         .get_rxfh_indir_size = netvsc_rss_indir_size,
1176         .get_rxfh       = netvsc_get_rxfh,
1177         .set_rxfh       = netvsc_set_rxfh,
1178 };
1179
1180 static const struct net_device_ops device_ops = {
1181         .ndo_open =                     netvsc_open,
1182         .ndo_stop =                     netvsc_close,
1183         .ndo_start_xmit =               netvsc_start_xmit,
1184         .ndo_set_rx_mode =              netvsc_set_multicast_list,
1185         .ndo_change_mtu =               netvsc_change_mtu,
1186         .ndo_validate_addr =            eth_validate_addr,
1187         .ndo_set_mac_address =          netvsc_set_mac_addr,
1188         .ndo_select_queue =             netvsc_select_queue,
1189         .ndo_get_stats64 =              netvsc_get_stats64,
1190 #ifdef CONFIG_NET_POLL_CONTROLLER
1191         .ndo_poll_controller =          netvsc_poll_controller,
1192 #endif
1193 };
1194
1195 /*
1196  * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
1197  * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
1198  * present send GARP packet to network peers with netif_notify_peers().
1199  */
1200 static void netvsc_link_change(struct work_struct *w)
1201 {
1202         struct net_device_context *ndev_ctx =
1203                 container_of(w, struct net_device_context, dwork.work);
1204         struct hv_device *device_obj = ndev_ctx->device_ctx;
1205         struct net_device *net = hv_get_drvdata(device_obj);
1206         struct netvsc_device *net_device;
1207         struct rndis_device *rdev;
1208         struct netvsc_reconfig *event = NULL;
1209         bool notify = false, reschedule = false;
1210         unsigned long flags, next_reconfig, delay;
1211
1212         rtnl_lock();
1213         if (ndev_ctx->start_remove)
1214                 goto out_unlock;
1215
1216         net_device = ndev_ctx->nvdev;
1217         rdev = net_device->extension;
1218
1219         next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
1220         if (time_is_after_jiffies(next_reconfig)) {
1221                 /* link_watch only sends one notification with current state
1222                  * per second, avoid doing reconfig more frequently. Handle
1223                  * wrap around.
1224                  */
1225                 delay = next_reconfig - jiffies;
1226                 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
1227                 schedule_delayed_work(&ndev_ctx->dwork, delay);
1228                 goto out_unlock;
1229         }
1230         ndev_ctx->last_reconfig = jiffies;
1231
1232         spin_lock_irqsave(&ndev_ctx->lock, flags);
1233         if (!list_empty(&ndev_ctx->reconfig_events)) {
1234                 event = list_first_entry(&ndev_ctx->reconfig_events,
1235                                          struct netvsc_reconfig, list);
1236                 list_del(&event->list);
1237                 reschedule = !list_empty(&ndev_ctx->reconfig_events);
1238         }
1239         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1240
1241         if (!event)
1242                 goto out_unlock;
1243
1244         switch (event->event) {
1245                 /* Only the following events are possible due to the check in
1246                  * netvsc_linkstatus_callback()
1247                  */
1248         case RNDIS_STATUS_MEDIA_CONNECT:
1249                 if (rdev->link_state) {
1250                         rdev->link_state = false;
1251                         netif_carrier_on(net);
1252                         netif_tx_wake_all_queues(net);
1253                 } else {
1254                         notify = true;
1255                 }
1256                 kfree(event);
1257                 break;
1258         case RNDIS_STATUS_MEDIA_DISCONNECT:
1259                 if (!rdev->link_state) {
1260                         rdev->link_state = true;
1261                         netif_carrier_off(net);
1262                         netif_tx_stop_all_queues(net);
1263                 }
1264                 kfree(event);
1265                 break;
1266         case RNDIS_STATUS_NETWORK_CHANGE:
1267                 /* Only makes sense if carrier is present */
1268                 if (!rdev->link_state) {
1269                         rdev->link_state = true;
1270                         netif_carrier_off(net);
1271                         netif_tx_stop_all_queues(net);
1272                         event->event = RNDIS_STATUS_MEDIA_CONNECT;
1273                         spin_lock_irqsave(&ndev_ctx->lock, flags);
1274                         list_add(&event->list, &ndev_ctx->reconfig_events);
1275                         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1276                         reschedule = true;
1277                 }
1278                 break;
1279         }
1280
1281         rtnl_unlock();
1282
1283         if (notify)
1284                 netdev_notify_peers(net);
1285
1286         /* link_watch only sends one notification with current state per
1287          * second, handle next reconfig event in 2 seconds.
1288          */
1289         if (reschedule)
1290                 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1291
1292         return;
1293
1294 out_unlock:
1295         rtnl_unlock();
1296 }
1297
1298 static struct net_device *get_netvsc_bymac(const u8 *mac)
1299 {
1300         struct net_device *dev;
1301
1302         ASSERT_RTNL();
1303
1304         for_each_netdev(&init_net, dev) {
1305                 if (dev->netdev_ops != &device_ops)
1306                         continue;       /* not a netvsc device */
1307
1308                 if (ether_addr_equal(mac, dev->perm_addr))
1309                         return dev;
1310         }
1311
1312         return NULL;
1313 }
1314
1315 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
1316 {
1317         struct net_device *dev;
1318
1319         ASSERT_RTNL();
1320
1321         for_each_netdev(&init_net, dev) {
1322                 struct net_device_context *net_device_ctx;
1323
1324                 if (dev->netdev_ops != &device_ops)
1325                         continue;       /* not a netvsc device */
1326
1327                 net_device_ctx = netdev_priv(dev);
1328                 if (net_device_ctx->nvdev == NULL)
1329                         continue;       /* device is removed */
1330
1331                 if (rtnl_dereference(net_device_ctx->vf_netdev) == vf_netdev)
1332                         return dev;     /* a match */
1333         }
1334
1335         return NULL;
1336 }
1337
1338 static int netvsc_register_vf(struct net_device *vf_netdev)
1339 {
1340         struct net_device *ndev;
1341         struct net_device_context *net_device_ctx;
1342         struct netvsc_device *netvsc_dev;
1343
1344         if (vf_netdev->addr_len != ETH_ALEN)
1345                 return NOTIFY_DONE;
1346
1347         /*
1348          * We will use the MAC address to locate the synthetic interface to
1349          * associate with the VF interface. If we don't find a matching
1350          * synthetic interface, move on.
1351          */
1352         ndev = get_netvsc_bymac(vf_netdev->perm_addr);
1353         if (!ndev)
1354                 return NOTIFY_DONE;
1355
1356         net_device_ctx = netdev_priv(ndev);
1357         netvsc_dev = net_device_ctx->nvdev;
1358         if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
1359                 return NOTIFY_DONE;
1360
1361         netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
1362         /*
1363          * Take a reference on the module.
1364          */
1365         try_module_get(THIS_MODULE);
1366
1367         dev_hold(vf_netdev);
1368         rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
1369         return NOTIFY_OK;
1370 }
1371
1372 static int netvsc_vf_up(struct net_device *vf_netdev)
1373 {
1374         struct net_device *ndev;
1375         struct netvsc_device *netvsc_dev;
1376         struct net_device_context *net_device_ctx;
1377
1378         ndev = get_netvsc_byref(vf_netdev);
1379         if (!ndev)
1380                 return NOTIFY_DONE;
1381
1382         net_device_ctx = netdev_priv(ndev);
1383         netvsc_dev = net_device_ctx->nvdev;
1384
1385         netdev_info(ndev, "VF up: %s\n", vf_netdev->name);
1386
1387         /*
1388          * Open the device before switching data path.
1389          */
1390         rndis_filter_open(netvsc_dev);
1391
1392         /*
1393          * notify the host to switch the data path.
1394          */
1395         netvsc_switch_datapath(ndev, true);
1396         netdev_info(ndev, "Data path switched to VF: %s\n", vf_netdev->name);
1397
1398         netif_carrier_off(ndev);
1399
1400         /* Now notify peers through VF device. */
1401         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, vf_netdev);
1402
1403         return NOTIFY_OK;
1404 }
1405
1406 static int netvsc_vf_down(struct net_device *vf_netdev)
1407 {
1408         struct net_device *ndev;
1409         struct netvsc_device *netvsc_dev;
1410         struct net_device_context *net_device_ctx;
1411
1412         ndev = get_netvsc_byref(vf_netdev);
1413         if (!ndev)
1414                 return NOTIFY_DONE;
1415
1416         net_device_ctx = netdev_priv(ndev);
1417         netvsc_dev = net_device_ctx->nvdev;
1418
1419         netdev_info(ndev, "VF down: %s\n", vf_netdev->name);
1420         netvsc_switch_datapath(ndev, false);
1421         netdev_info(ndev, "Data path switched from VF: %s\n", vf_netdev->name);
1422         rndis_filter_close(netvsc_dev);
1423         netif_carrier_on(ndev);
1424
1425         /* Now notify peers through netvsc device. */
1426         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, ndev);
1427
1428         return NOTIFY_OK;
1429 }
1430
1431 static int netvsc_unregister_vf(struct net_device *vf_netdev)
1432 {
1433         struct net_device *ndev;
1434         struct net_device_context *net_device_ctx;
1435
1436         ndev = get_netvsc_byref(vf_netdev);
1437         if (!ndev)
1438                 return NOTIFY_DONE;
1439
1440         net_device_ctx = netdev_priv(ndev);
1441
1442         netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
1443
1444         RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
1445         dev_put(vf_netdev);
1446         module_put(THIS_MODULE);
1447         return NOTIFY_OK;
1448 }
1449
1450 static int netvsc_probe(struct hv_device *dev,
1451                         const struct hv_vmbus_device_id *dev_id)
1452 {
1453         struct net_device *net = NULL;
1454         struct net_device_context *net_device_ctx;
1455         struct netvsc_device_info device_info;
1456         struct netvsc_device *nvdev;
1457         int ret;
1458
1459         net = alloc_etherdev_mq(sizeof(struct net_device_context),
1460                                 VRSS_CHANNEL_MAX);
1461         if (!net)
1462                 return -ENOMEM;
1463
1464         netif_carrier_off(net);
1465
1466         netvsc_init_settings(net);
1467
1468         net_device_ctx = netdev_priv(net);
1469         net_device_ctx->device_ctx = dev;
1470         net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
1471         if (netif_msg_probe(net_device_ctx))
1472                 netdev_dbg(net, "netvsc msg_enable: %d\n",
1473                            net_device_ctx->msg_enable);
1474
1475         hv_set_drvdata(dev, net);
1476
1477         net_device_ctx->start_remove = false;
1478
1479         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
1480         INIT_WORK(&net_device_ctx->work, do_set_multicast);
1481
1482         spin_lock_init(&net_device_ctx->lock);
1483         INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
1484
1485         net->netdev_ops = &device_ops;
1486         net->ethtool_ops = &ethtool_ops;
1487         SET_NETDEV_DEV(net, &dev->device);
1488
1489         /* We always need headroom for rndis header */
1490         net->needed_headroom = RNDIS_AND_PPI_SIZE;
1491
1492         /* Notify the netvsc driver of the new device */
1493         memset(&device_info, 0, sizeof(device_info));
1494         device_info.ring_size = ring_size;
1495         device_info.max_num_vrss_chns = min_t(u32, VRSS_CHANNEL_DEFAULT,
1496                                               num_online_cpus());
1497         ret = rndis_filter_device_add(dev, &device_info);
1498         if (ret != 0) {
1499                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
1500                 free_netdev(net);
1501                 hv_set_drvdata(dev, NULL);
1502                 return ret;
1503         }
1504         memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
1505
1506         /* hw_features computed in rndis_filter_device_add */
1507         net->features = net->hw_features |
1508                 NETIF_F_HIGHDMA | NETIF_F_SG |
1509                 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
1510         net->vlan_features = net->features;
1511
1512         nvdev = net_device_ctx->nvdev;
1513         netif_set_real_num_tx_queues(net, nvdev->num_chn);
1514         netif_set_real_num_rx_queues(net, nvdev->num_chn);
1515
1516         /* MTU range: 68 - 1500 or 65521 */
1517         net->min_mtu = NETVSC_MTU_MIN;
1518         if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
1519                 net->max_mtu = NETVSC_MTU - ETH_HLEN;
1520         else
1521                 net->max_mtu = ETH_DATA_LEN;
1522
1523         ret = register_netdev(net);
1524         if (ret != 0) {
1525                 pr_err("Unable to register netdev.\n");
1526                 rndis_filter_device_remove(dev, nvdev);
1527                 free_netdev(net);
1528         }
1529
1530         return ret;
1531 }
1532
1533 static int netvsc_remove(struct hv_device *dev)
1534 {
1535         struct net_device *net;
1536         struct net_device_context *ndev_ctx;
1537
1538         net = hv_get_drvdata(dev);
1539
1540         if (net == NULL) {
1541                 dev_err(&dev->device, "No net device to remove\n");
1542                 return 0;
1543         }
1544
1545         ndev_ctx = netdev_priv(net);
1546
1547         /* Avoid racing with netvsc_change_mtu()/netvsc_set_channels()
1548          * removing the device.
1549          */
1550         rtnl_lock();
1551         ndev_ctx->start_remove = true;
1552         rtnl_unlock();
1553
1554         cancel_delayed_work_sync(&ndev_ctx->dwork);
1555         cancel_work_sync(&ndev_ctx->work);
1556
1557         /* Stop outbound asap */
1558         netif_tx_disable(net);
1559
1560         unregister_netdev(net);
1561
1562         /*
1563          * Call to the vsc driver to let it know that the device is being
1564          * removed
1565          */
1566         rndis_filter_device_remove(dev, ndev_ctx->nvdev);
1567
1568         hv_set_drvdata(dev, NULL);
1569
1570         free_netdev(net);
1571         return 0;
1572 }
1573
1574 static const struct hv_vmbus_device_id id_table[] = {
1575         /* Network guid */
1576         { HV_NIC_GUID, },
1577         { },
1578 };
1579
1580 MODULE_DEVICE_TABLE(vmbus, id_table);
1581
1582 /* The one and only one */
1583 static struct  hv_driver netvsc_drv = {
1584         .name = KBUILD_MODNAME,
1585         .id_table = id_table,
1586         .probe = netvsc_probe,
1587         .remove = netvsc_remove,
1588 };
1589
1590 /*
1591  * On Hyper-V, every VF interface is matched with a corresponding
1592  * synthetic interface. The synthetic interface is presented first
1593  * to the guest. When the corresponding VF instance is registered,
1594  * we will take care of switching the data path.
1595  */
1596 static int netvsc_netdev_event(struct notifier_block *this,
1597                                unsigned long event, void *ptr)
1598 {
1599         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
1600
1601         /* Skip our own events */
1602         if (event_dev->netdev_ops == &device_ops)
1603                 return NOTIFY_DONE;
1604
1605         /* Avoid non-Ethernet type devices */
1606         if (event_dev->type != ARPHRD_ETHER)
1607                 return NOTIFY_DONE;
1608
1609         /* Avoid Vlan dev with same MAC registering as VF */
1610         if (is_vlan_dev(event_dev))
1611                 return NOTIFY_DONE;
1612
1613         /* Avoid Bonding master dev with same MAC registering as VF */
1614         if ((event_dev->priv_flags & IFF_BONDING) &&
1615             (event_dev->flags & IFF_MASTER))
1616                 return NOTIFY_DONE;
1617
1618         switch (event) {
1619         case NETDEV_REGISTER:
1620                 return netvsc_register_vf(event_dev);
1621         case NETDEV_UNREGISTER:
1622                 return netvsc_unregister_vf(event_dev);
1623         case NETDEV_UP:
1624                 return netvsc_vf_up(event_dev);
1625         case NETDEV_DOWN:
1626                 return netvsc_vf_down(event_dev);
1627         default:
1628                 return NOTIFY_DONE;
1629         }
1630 }
1631
1632 static struct notifier_block netvsc_netdev_notifier = {
1633         .notifier_call = netvsc_netdev_event,
1634 };
1635
1636 static void __exit netvsc_drv_exit(void)
1637 {
1638         unregister_netdevice_notifier(&netvsc_netdev_notifier);
1639         vmbus_driver_unregister(&netvsc_drv);
1640 }
1641
1642 static int __init netvsc_drv_init(void)
1643 {
1644         int ret;
1645
1646         if (ring_size < RING_SIZE_MIN) {
1647                 ring_size = RING_SIZE_MIN;
1648                 pr_info("Increased ring_size to %d (min allowed)\n",
1649                         ring_size);
1650         }
1651         ret = vmbus_driver_register(&netvsc_drv);
1652
1653         if (ret)
1654                 return ret;
1655
1656         register_netdevice_notifier(&netvsc_netdev_notifier);
1657         return 0;
1658 }
1659
1660 MODULE_LICENSE("GPL");
1661 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
1662
1663 module_init(netvsc_drv_init);
1664 module_exit(netvsc_drv_exit);