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 #define NETVSC_HW_FEATURES      (NETIF_F_RXCSUM | \
  46                                  NETIF_F_SG | \
  47                                  NETIF_F_TSO | \
  48                                  NETIF_F_TSO6 | \
  49                                  NETIF_F_HW_CSUM)
  50 static int ring_size = 128;
  51 module_param(ring_size, int, S_IRUGO);
  52 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
  53
  54 static int max_num_vrss_chns = 8;
  55
  56 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
  57                                 NETIF_MSG_LINK | NETIF_MSG_IFUP |
  58                                 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
  59                                 NETIF_MSG_TX_ERR;
  60
  61 static int debug = -1;
  62 module_param(debug, int, S_IRUGO);
  63 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  64
  65 static void do_set_multicast(struct work_struct *w)
  66 {
  67         struct net_device_context *ndevctx =
  68                 container_of(w, struct net_device_context, work);
  69         struct hv_device *device_obj = ndevctx->device_ctx;
  70         struct net_device *ndev = hv_get_drvdata(device_obj);
  71         struct netvsc_device *nvdev = ndevctx->nvdev;
  72         struct rndis_device *rdev;
  73
  74         if (!nvdev)
  75                 return;
  76
  77         rdev = nvdev->extension;
  78         if (rdev == NULL)
  79                 return;
  80
  81         if (ndev->flags & IFF_PROMISC)
  82                 rndis_filter_set_packet_filter(rdev,
  83                         NDIS_PACKET_TYPE_PROMISCUOUS);
  84         else
  85                 rndis_filter_set_packet_filter(rdev,
  86                         NDIS_PACKET_TYPE_BROADCAST |
  87                         NDIS_PACKET_TYPE_ALL_MULTICAST |
  88                         NDIS_PACKET_TYPE_DIRECTED);
  89 }
  90
  91 static void netvsc_set_multicast_list(struct net_device *net)
  92 {
  93         struct net_device_context *net_device_ctx = netdev_priv(net);
  94
  95         schedule_work(&net_device_ctx->work);
  96 }
  97
  98 static int netvsc_open(struct net_device *net)
  99 {
 100         struct netvsc_device *nvdev = net_device_to_netvsc_device(net);
 101         struct rndis_device *rdev;
 102         int ret = 0;
 103
 104         netif_carrier_off(net);
 105
 106         /* Open up the device */
 107         ret = rndis_filter_open(nvdev);
 108         if (ret != 0) {
 109                 netdev_err(net, "unable to open device (ret %d).\n", ret);
 110                 return ret;
 111         }
 112
 113         netif_tx_wake_all_queues(net);
 114
 115         rdev = nvdev->extension;
 116         if (!rdev->link_state)
 117                 netif_carrier_on(net);
 118
 119         return ret;
 120 }
 121
 122 static int netvsc_close(struct net_device *net)
 123 {
 124         struct net_device_context *net_device_ctx = netdev_priv(net);
 125         struct netvsc_device *nvdev = net_device_ctx->nvdev;
 126         int ret;
 127         u32 aread, awrite, i, msec = 10, retry = 0, retry_max = 20;
 128         struct vmbus_channel *chn;
 129
 130         netif_tx_disable(net);
 131
 132         /* Make sure netvsc_set_multicast_list doesn't re-enable filter! */
 133         cancel_work_sync(&net_device_ctx->work);
 134         ret = rndis_filter_close(nvdev);
 135         if (ret != 0) {
 136                 netdev_err(net, "unable to close device (ret %d).\n", ret);
 137                 return ret;
 138         }
 139
 140         /* Ensure pending bytes in ring are read */
 141         while (true) {
 142                 aread = 0;
 143                 for (i = 0; i < nvdev->num_chn; i++) {
 144                         chn = nvdev->chn_table[i];
 145                         if (!chn)
 146                                 continue;
 147
 148                         hv_get_ringbuffer_availbytes(&chn->inbound, &aread,
 149                                                      &awrite);
 150
 151                         if (aread)
 152                                 break;
 153
 154                         hv_get_ringbuffer_availbytes(&chn->outbound, &aread,
 155                                                      &awrite);
 156
 157                         if (aread)
 158                                 break;
 159                 }
 160
 161                 retry++;
 162                 if (retry > retry_max || aread == 0)
 163                         break;
 164
 165                 msleep(msec);
 166
 167                 if (msec < 1000)
 168                         msec *= 2;
 169         }
 170
 171         if (aread) {
 172                 netdev_err(net, "Ring buffer not empty after closing rndis\n");
 173                 ret = -ETIMEDOUT;
 174         }
 175
 176         return ret;
 177 }
 178
 179 static void *init_ppi_data(struct rndis_message *msg, u32 ppi_size,
 180                                 int pkt_type)
 181 {
 182         struct rndis_packet *rndis_pkt;
 183         struct rndis_per_packet_info *ppi;
 184
 185         rndis_pkt = &msg->msg.pkt;
 186         rndis_pkt->data_offset += ppi_size;
 187
 188         ppi = (struct rndis_per_packet_info *)((void *)rndis_pkt +
 189                 rndis_pkt->per_pkt_info_offset + rndis_pkt->per_pkt_info_len);
 190
 191         ppi->size = ppi_size;
 192         ppi->type = pkt_type;
 193         ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
 194
 195         rndis_pkt->per_pkt_info_len += ppi_size;
 196
 197         return ppi;
 198 }
 199
 200 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
 201                         void *accel_priv, select_queue_fallback_t fallback)
 202 {
 203         struct net_device_context *net_device_ctx = netdev_priv(ndev);
 204         struct netvsc_device *nvsc_dev = net_device_ctx->nvdev;
 205         u32 hash;
 206         u16 q_idx = 0;
 207
 208         if (nvsc_dev == NULL || ndev->real_num_tx_queues <= 1)
 209                 return 0;
 210
 211         hash = skb_get_hash(skb);
 212         q_idx = nvsc_dev->send_table[hash % VRSS_SEND_TAB_SIZE] %
 213                 ndev->real_num_tx_queues;
 214
 215         if (!nvsc_dev->chn_table[q_idx])
 216                 q_idx = 0;
 217
 218         return q_idx;
 219 }
 220
 221 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
 222                         struct hv_page_buffer *pb)
 223 {
 224         int j = 0;
 225
 226         /* Deal with compund pages by ignoring unused part
 227          * of the page.
 228          */
 229         page += (offset >> PAGE_SHIFT);
 230         offset &= ~PAGE_MASK;
 231
 232         while (len > 0) {
 233                 unsigned long bytes;
 234
 235                 bytes = PAGE_SIZE - offset;
 236                 if (bytes > len)
 237                         bytes = len;
 238                 pb[j].pfn = page_to_pfn(page);
 239                 pb[j].offset = offset;
 240                 pb[j].len = bytes;
 241
 242                 offset += bytes;
 243                 len -= bytes;
 244
 245                 if (offset == PAGE_SIZE && len) {
 246                         page++;
 247                         offset = 0;
 248                         j++;
 249                 }
 250         }
 251
 252         return j + 1;
 253 }
 254
 255 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
 256                            struct hv_netvsc_packet *packet,
 257                            struct hv_page_buffer **page_buf)
 258 {
 259         struct hv_page_buffer *pb = *page_buf;
 260         u32 slots_used = 0;
 261         char *data = skb->data;
 262         int frags = skb_shinfo(skb)->nr_frags;
 263         int i;
 264
 265         /* The packet is laid out thus:
 266          * 1. hdr: RNDIS header and PPI
 267          * 2. skb linear data
 268          * 3. skb fragment data
 269          */
 270         if (hdr != NULL)
 271                 slots_used += fill_pg_buf(virt_to_page(hdr),
 272                                         offset_in_page(hdr),
 273                                         len, &pb[slots_used]);
 274
 275         packet->rmsg_size = len;
 276         packet->rmsg_pgcnt = slots_used;
 277
 278         slots_used += fill_pg_buf(virt_to_page(data),
 279                                 offset_in_page(data),
 280                                 skb_headlen(skb), &pb[slots_used]);
 281
 282         for (i = 0; i < frags; i++) {
 283                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 284
 285                 slots_used += fill_pg_buf(skb_frag_page(frag),
 286                                         frag->page_offset,
 287                                         skb_frag_size(frag), &pb[slots_used]);
 288         }
 289         return slots_used;
 290 }
 291
 292 static int count_skb_frag_slots(struct sk_buff *skb)
 293 {
 294         int i, frags = skb_shinfo(skb)->nr_frags;
 295         int pages = 0;
 296
 297         for (i = 0; i < frags; i++) {
 298                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 299                 unsigned long size = skb_frag_size(frag);
 300                 unsigned long offset = frag->page_offset;
 301
 302                 /* Skip unused frames from start of page */
 303                 offset &= ~PAGE_MASK;
 304                 pages += PFN_UP(offset + size);
 305         }
 306         return pages;
 307 }
 308
 309 static int netvsc_get_slots(struct sk_buff *skb)
 310 {
 311         char *data = skb->data;
 312         unsigned int offset = offset_in_page(data);
 313         unsigned int len = skb_headlen(skb);
 314         int slots;
 315         int frag_slots;
 316
 317         slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
 318         frag_slots = count_skb_frag_slots(skb);
 319         return slots + frag_slots;
 320 }
 321
 322 static u32 get_net_transport_info(struct sk_buff *skb, u32 *trans_off)
 323 {
 324         u32 ret_val = TRANSPORT_INFO_NOT_IP;
 325
 326         if ((eth_hdr(skb)->h_proto != htons(ETH_P_IP)) &&
 327                 (eth_hdr(skb)->h_proto != htons(ETH_P_IPV6))) {
 328                 goto not_ip;
 329         }
 330
 331         *trans_off = skb_transport_offset(skb);
 332
 333         if ((eth_hdr(skb)->h_proto == htons(ETH_P_IP))) {
 334                 struct iphdr *iphdr = ip_hdr(skb);
 335
 336                 if (iphdr->protocol == IPPROTO_TCP)
 337                         ret_val = TRANSPORT_INFO_IPV4_TCP;
 338                 else if (iphdr->protocol == IPPROTO_UDP)
 339                         ret_val = TRANSPORT_INFO_IPV4_UDP;
 340         } else {
 341                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
 342                         ret_val = TRANSPORT_INFO_IPV6_TCP;
 343                 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
 344                         ret_val = TRANSPORT_INFO_IPV6_UDP;
 345         }
 346
 347 not_ip:
 348         return ret_val;
 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         struct ndis_tcp_ip_checksum_info *csum_info;
 362         int  hdr_offset;
 363         u32 net_trans_info;
 364         u32 hash;
 365         u32 skb_length;
 366         struct hv_page_buffer page_buf[MAX_PAGE_BUFFER_COUNT];
 367         struct hv_page_buffer *pb = page_buf;
 368
 369         /* We will atmost need two pages to describe the rndis
 370          * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
 371          * of pages in a single packet. If skb is scattered around
 372          * more pages we try linearizing it.
 373          */
 374
 375         skb_length = skb->len;
 376         num_data_pgs = netvsc_get_slots(skb) + 2;
 377
 378         if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
 379                 ++net_device_ctx->eth_stats.tx_scattered;
 380
 381                 if (skb_linearize(skb))
 382                         goto no_memory;
 383
 384                 num_data_pgs = netvsc_get_slots(skb) + 2;
 385                 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
 386                         ++net_device_ctx->eth_stats.tx_too_big;
 387                         goto drop;
 388                 }
 389         }
 390
 391         /*
 392          * Place the rndis header in the skb head room and
 393          * the skb->cb will be used for hv_netvsc_packet
 394          * structure.
 395          */
 396         ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
 397         if (ret)
 398                 goto no_memory;
 399
 400         /* Use the skb control buffer for building up the packet */
 401         BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
 402                         FIELD_SIZEOF(struct sk_buff, cb));
 403         packet = (struct hv_netvsc_packet *)skb->cb;
 404
 405         packet->q_idx = skb_get_queue_mapping(skb);
 406
 407         packet->total_data_buflen = skb->len;
 408
 409         rndis_msg = (struct rndis_message *)skb->head;
 410
 411         memset(rndis_msg, 0, RNDIS_AND_PPI_SIZE);
 412
 413         /* Add the rndis header */
 414         rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
 415         rndis_msg->msg_len = packet->total_data_buflen;
 416         rndis_pkt = &rndis_msg->msg.pkt;
 417         rndis_pkt->data_offset = sizeof(struct rndis_packet);
 418         rndis_pkt->data_len = packet->total_data_buflen;
 419         rndis_pkt->per_pkt_info_offset = sizeof(struct rndis_packet);
 420
 421         rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
 422
 423         hash = skb_get_hash_raw(skb);
 424         if (hash != 0 && net->real_num_tx_queues > 1) {
 425                 rndis_msg_size += NDIS_HASH_PPI_SIZE;
 426                 ppi = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
 427                                     NBL_HASH_VALUE);
 428                 *(u32 *)((void *)ppi + ppi->ppi_offset) = hash;
 429         }
 430
 431         if (skb_vlan_tag_present(skb)) {
 432                 struct ndis_pkt_8021q_info *vlan;
 433
 434                 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
 435                 ppi = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
 436                                         IEEE_8021Q_INFO);
 437                 vlan = (struct ndis_pkt_8021q_info *)((void *)ppi +
 438                                                 ppi->ppi_offset);
 439                 vlan->vlanid = skb->vlan_tci & VLAN_VID_MASK;
 440                 vlan->pri = (skb->vlan_tci & VLAN_PRIO_MASK) >>
 441                                 VLAN_PRIO_SHIFT;
 442         }
 443
 444         net_trans_info = get_net_transport_info(skb, &hdr_offset);
 445         if (net_trans_info == TRANSPORT_INFO_NOT_IP)
 446                 goto do_send;
 447
 448         /*
 449          * Setup the sendside checksum offload only if this is not a
 450          * GSO packet.
 451          */
 452         if (skb_is_gso(skb)) {
 453                 struct ndis_tcp_lso_info *lso_info;
 454
 455                 rndis_msg_size += NDIS_LSO_PPI_SIZE;
 456                 ppi = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
 457                                     TCP_LARGESEND_PKTINFO);
 458
 459                 lso_info = (struct ndis_tcp_lso_info *)((void *)ppi +
 460                                                         ppi->ppi_offset);
 461
 462                 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
 463                 if (net_trans_info & (INFO_IPV4 << 16)) {
 464                         lso_info->lso_v2_transmit.ip_version =
 465                                 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
 466                         ip_hdr(skb)->tot_len = 0;
 467                         ip_hdr(skb)->check = 0;
 468                         tcp_hdr(skb)->check =
 469                                 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
 470                                                    ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 471                 } else {
 472                         lso_info->lso_v2_transmit.ip_version =
 473                                 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
 474                         ipv6_hdr(skb)->payload_len = 0;
 475                         tcp_hdr(skb)->check =
 476                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 477                                                  &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 478                 }
 479                 lso_info->lso_v2_transmit.tcp_header_offset = hdr_offset;
 480                 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
 481                 goto do_send;
 482         }
 483
 484         if ((skb->ip_summed == CHECKSUM_NONE) ||
 485             (skb->ip_summed == CHECKSUM_UNNECESSARY))
 486                 goto do_send;
 487
 488         rndis_msg_size += NDIS_CSUM_PPI_SIZE;
 489         ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
 490                             TCPIP_CHKSUM_PKTINFO);
 491
 492         csum_info = (struct ndis_tcp_ip_checksum_info *)((void *)ppi +
 493                         ppi->ppi_offset);
 494
 495         if (net_trans_info & (INFO_IPV4 << 16))
 496                 csum_info->transmit.is_ipv4 = 1;
 497         else
 498                 csum_info->transmit.is_ipv6 = 1;
 499
 500         if (net_trans_info & INFO_TCP) {
 501                 csum_info->transmit.tcp_checksum = 1;
 502                 csum_info->transmit.tcp_header_offset = hdr_offset;
 503         } else if (net_trans_info & INFO_UDP) {
 504                 /* UDP checksum offload is not supported on ws2008r2.
 505                  * Furthermore, on ws2012 and ws2012r2, there are some
 506                  * issues with udp checksum offload from Linux guests.
 507                  * (these are host issues).
 508                  * For now compute the checksum here.
 509                  */
 510                 struct udphdr *uh;
 511                 u16 udp_len;
 512
 513                 ret = skb_cow_head(skb, 0);
 514                 if (ret)
 515                         goto no_memory;
 516
 517                 uh = udp_hdr(skb);
 518                 udp_len = ntohs(uh->len);
 519                 uh->check = 0;
 520                 uh->check = csum_tcpudp_magic(ip_hdr(skb)->saddr,
 521                                               ip_hdr(skb)->daddr,
 522                                               udp_len, IPPROTO_UDP,
 523                                               csum_partial(uh, udp_len, 0));
 524                 if (uh->check == 0)
 525                         uh->check = CSUM_MANGLED_0;
 526
 527                 csum_info->transmit.udp_checksum = 0;
 528         }
 529
 530 do_send:
 531         /* Start filling in the page buffers with the rndis hdr */
 532         rndis_msg->msg_len += rndis_msg_size;
 533         packet->total_data_buflen = rndis_msg->msg_len;
 534         packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
 535                                                skb, packet, &pb);
 536
 537         /* timestamp packet in software */
 538         skb_tx_timestamp(skb);
 539         ret = netvsc_send(net_device_ctx->device_ctx, packet,
 540                           rndis_msg, &pb, skb);
 541         if (likely(ret == 0)) {
 542                 struct netvsc_stats *tx_stats = this_cpu_ptr(net_device_ctx->tx_stats);
 543
 544                 u64_stats_update_begin(&tx_stats->syncp);
 545                 tx_stats->packets++;
 546                 tx_stats->bytes += skb_length;
 547                 u64_stats_update_end(&tx_stats->syncp);
 548                 return NETDEV_TX_OK;
 549         }
 550
 551         if (ret == -EAGAIN) {
 552                 ++net_device_ctx->eth_stats.tx_busy;
 553                 return NETDEV_TX_BUSY;
 554         }
 555
 556         if (ret == -ENOSPC)
 557                 ++net_device_ctx->eth_stats.tx_no_space;
 558
 559 drop:
 560         dev_kfree_skb_any(skb);
 561         net->stats.tx_dropped++;
 562
 563         return NETDEV_TX_OK;
 564
 565 no_memory:
 566         ++net_device_ctx->eth_stats.tx_no_memory;
 567         goto drop;
 568 }
 569
 570 /*
 571  * netvsc_linkstatus_callback - Link up/down notification
 572  */
 573 void netvsc_linkstatus_callback(struct hv_device *device_obj,
 574                                 struct rndis_message *resp)
 575 {
 576         struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
 577         struct net_device *net;
 578         struct net_device_context *ndev_ctx;
 579         struct netvsc_reconfig *event;
 580         unsigned long flags;
 581
 582         net = hv_get_drvdata(device_obj);
 583
 584         if (!net)
 585                 return;
 586
 587         ndev_ctx = netdev_priv(net);
 588
 589         /* Update the physical link speed when changing to another vSwitch */
 590         if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
 591                 u32 speed;
 592
 593                 speed = *(u32 *)((void *)indicate + indicate->
 594                                  status_buf_offset) / 10000;
 595                 ndev_ctx->speed = speed;
 596                 return;
 597         }
 598
 599         /* Handle these link change statuses below */
 600         if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
 601             indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
 602             indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
 603                 return;
 604
 605         if (net->reg_state != NETREG_REGISTERED)
 606                 return;
 607
 608         event = kzalloc(sizeof(*event), GFP_ATOMIC);
 609         if (!event)
 610                 return;
 611         event->event = indicate->status;
 612
 613         spin_lock_irqsave(&ndev_ctx->lock, flags);
 614         list_add_tail(&event->list, &ndev_ctx->reconfig_events);
 615         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
 616
 617         schedule_delayed_work(&ndev_ctx->dwork, 0);
 618 }
 619
 620 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
 621                                 struct hv_netvsc_packet *packet,
 622                                 struct ndis_tcp_ip_checksum_info *csum_info,
 623                                 void *data, u16 vlan_tci)
 624 {
 625         struct sk_buff *skb;
 626
 627         skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
 628         if (!skb)
 629                 return skb;
 630
 631         /*
 632          * Copy to skb. This copy is needed here since the memory pointed by
 633          * hv_netvsc_packet cannot be deallocated
 634          */
 635         memcpy(skb_put(skb, packet->total_data_buflen), data,
 636                packet->total_data_buflen);
 637
 638         skb->protocol = eth_type_trans(skb, net);
 639         if (csum_info) {
 640                 /* We only look at the IP checksum here.
 641                  * Should we be dropping the packet if checksum
 642                  * failed? How do we deal with other checksums - TCP/UDP?
 643                  */
 644                 if (csum_info->receive.ip_checksum_succeeded)
 645                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 646                 else
 647                         skb->ip_summed = CHECKSUM_NONE;
 648         }
 649
 650         if (vlan_tci & VLAN_TAG_PRESENT)
 651                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 652                                        vlan_tci);
 653
 654         return skb;
 655 }
 656
 657 /*
 658  * netvsc_recv_callback -  Callback when we receive a packet from the
 659  * "wire" on the specified device.
 660  */
 661 int netvsc_recv_callback(struct hv_device *device_obj,
 662                                 struct hv_netvsc_packet *packet,
 663                                 void **data,
 664                                 struct ndis_tcp_ip_checksum_info *csum_info,
 665                                 struct vmbus_channel *channel,
 666                                 u16 vlan_tci)
 667 {
 668         struct net_device *net = hv_get_drvdata(device_obj);
 669         struct net_device_context *net_device_ctx = netdev_priv(net);
 670         struct net_device *vf_netdev;
 671         struct sk_buff *skb;
 672         struct netvsc_stats *rx_stats;
 673         u32 bytes_recvd = packet->total_data_buflen;
 674         int ret = 0;
 675
 676         if (!net || net->reg_state != NETREG_REGISTERED)
 677                 return NVSP_STAT_FAIL;
 678
 679         vf_netdev = rcu_dereference(net_device_ctx->vf_netdev);
 680         if (vf_netdev) {
 681                 struct sk_buff *vf_skb;
 682
 683                 atomic_inc(&net_device_ctx->vf_use_cnt);
 684                 if (!net_device_ctx->vf_inject) {
 685                         /*
 686                          * We raced; just move on.
 687                          */
 688                         atomic_dec(&net_device_ctx->vf_use_cnt);
 689                         goto vf_injection_done;
 690                 }
 691
 692                 /*
 693                  * Inject this packet into the VF inerface.
 694                  * On Hyper-V, multicast and brodcast packets
 695                  * are only delivered on the synthetic interface
 696                  * (after subjecting these to policy filters on
 697                  * the host). Deliver these via the VF interface
 698                  * in the guest.
 699                  */
 700                 vf_skb = netvsc_alloc_recv_skb(vf_netdev,
 701                                                packet, csum_info, *data,
 702                                                vlan_tci);
 703                 if (vf_skb != NULL) {
 704                         ++vf_netdev->stats.rx_packets;
 705                         vf_netdev->stats.rx_bytes += bytes_recvd;
 706                         netif_receive_skb(vf_skb);
 707                 } else {
 708                         ++net->stats.rx_dropped;
 709                         ret = NVSP_STAT_FAIL;
 710                 }
 711                 atomic_dec(&net_device_ctx->vf_use_cnt);
 712                 return ret;
 713         }
 714
 715 vf_injection_done:
 716         rx_stats = this_cpu_ptr(net_device_ctx->rx_stats);
 717
 718         /* Allocate a skb - TODO direct I/O to pages? */
 719         skb = netvsc_alloc_recv_skb(net, packet, csum_info, *data, vlan_tci);
 720         if (unlikely(!skb)) {
 721                 ++net->stats.rx_dropped;
 722                 return NVSP_STAT_FAIL;
 723         }
 724         skb_record_rx_queue(skb, channel->
 725                             offermsg.offer.sub_channel_index);
 726
 727         u64_stats_update_begin(&rx_stats->syncp);
 728         rx_stats->packets++;
 729         rx_stats->bytes += packet->total_data_buflen;
 730         u64_stats_update_end(&rx_stats->syncp);
 731
 732         /*
 733          * Pass the skb back up. Network stack will deallocate the skb when it
 734          * is done.
 735          * TODO - use NAPI?
 736          */
 737         netif_rx(skb);
 738
 739         return 0;
 740 }
 741
 742 static void netvsc_get_drvinfo(struct net_device *net,
 743                                struct ethtool_drvinfo *info)
 744 {
 745         struct net_device_context *net_device_ctx = netdev_priv(net);
 746         struct hv_device *dev = net_device_ctx->device_ctx;
 747
 748         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
 749         strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
 750         strlcpy(info->bus_info, vmbus_dev_name(dev), sizeof(info->bus_info));
 751 }
 752
 753 static void netvsc_get_channels(struct net_device *net,
 754                                 struct ethtool_channels *channel)
 755 {
 756         struct net_device_context *net_device_ctx = netdev_priv(net);
 757         struct netvsc_device *nvdev = net_device_ctx->nvdev;
 758
 759         if (nvdev) {
 760                 channel->max_combined   = nvdev->max_chn;
 761                 channel->combined_count = nvdev->num_chn;
 762         }
 763 }
 764
 765 static int netvsc_set_channels(struct net_device *net,
 766                                struct ethtool_channels *channels)
 767 {
 768         struct net_device_context *net_device_ctx = netdev_priv(net);
 769         struct hv_device *dev = net_device_ctx->device_ctx;
 770         struct netvsc_device *nvdev = net_device_ctx->nvdev;
 771         struct netvsc_device_info device_info;
 772         u32 num_chn;
 773         u32 max_chn;
 774         int ret = 0;
 775         bool recovering = false;
 776
 777         if (net_device_ctx->start_remove || !nvdev || nvdev->destroy)
 778                 return -ENODEV;
 779
 780         num_chn = nvdev->num_chn;
 781         max_chn = min_t(u32, nvdev->max_chn, num_online_cpus());
 782
 783         if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5) {
 784                 pr_info("vRSS unsupported before NVSP Version 5\n");
 785                 return -EINVAL;
 786         }
 787
 788         /* We do not support rx, tx, or other */
 789         if (!channels ||
 790             channels->rx_count ||
 791             channels->tx_count ||
 792             channels->other_count ||
 793             (channels->combined_count < 1))
 794                 return -EINVAL;
 795
 796         if (channels->combined_count > max_chn) {
 797                 pr_info("combined channels too high, using %d\n", max_chn);
 798                 channels->combined_count = max_chn;
 799         }
 800
 801         ret = netvsc_close(net);
 802         if (ret)
 803                 goto out;
 804
 805  do_set:
 806         net_device_ctx->start_remove = true;
 807         rndis_filter_device_remove(dev);
 808
 809         nvdev->num_chn = channels->combined_count;
 810
 811         memset(&device_info, 0, sizeof(device_info));
 812         device_info.num_chn = nvdev->num_chn; /* passed to RNDIS */
 813         device_info.ring_size = ring_size;
 814         device_info.max_num_vrss_chns = max_num_vrss_chns;
 815
 816         ret = rndis_filter_device_add(dev, &device_info);
 817         if (ret) {
 818                 if (recovering) {
 819                         netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
 820                         return ret;
 821                 }
 822                 goto recover;
 823         }
 824
 825         nvdev = net_device_ctx->nvdev;
 826
 827         ret = netif_set_real_num_tx_queues(net, nvdev->num_chn);
 828         if (ret) {
 829                 if (recovering) {
 830                         netdev_err(net, "could not set tx queue count (ret %d)\n", ret);
 831                         return ret;
 832                 }
 833                 goto recover;
 834         }
 835
 836         ret = netif_set_real_num_rx_queues(net, nvdev->num_chn);
 837         if (ret) {
 838                 if (recovering) {
 839                         netdev_err(net, "could not set rx queue count (ret %d)\n", ret);
 840                         return ret;
 841                 }
 842                 goto recover;
 843         }
 844
 845  out:
 846         netvsc_open(net);
 847         net_device_ctx->start_remove = false;
 848         /* We may have missed link change notifications */
 849         schedule_delayed_work(&net_device_ctx->dwork, 0);
 850
 851         return ret;
 852
 853  recover:
 854         /* If the above failed, we attempt to recover through the same
 855          * process but with the original number of channels.
 856          */
 857         netdev_err(net, "could not set channels, recovering\n");
 858         recovering = true;
 859         channels->combined_count = num_chn;
 860         goto do_set;
 861 }
 862
 863 static bool netvsc_validate_ethtool_ss_cmd(const struct ethtool_cmd *cmd)
 864 {
 865         struct ethtool_cmd diff1 = *cmd;
 866         struct ethtool_cmd diff2 = {};
 867
 868         ethtool_cmd_speed_set(&diff1, 0);
 869         diff1.duplex = 0;
 870         /* advertising and cmd are usually set */
 871         diff1.advertising = 0;
 872         diff1.cmd = 0;
 873         /* We set port to PORT_OTHER */
 874         diff2.port = PORT_OTHER;
 875
 876         return !memcmp(&diff1, &diff2, sizeof(diff1));
 877 }
 878
 879 static void netvsc_init_settings(struct net_device *dev)
 880 {
 881         struct net_device_context *ndc = netdev_priv(dev);
 882
 883         ndc->speed = SPEED_UNKNOWN;
 884         ndc->duplex = DUPLEX_UNKNOWN;
 885 }
 886
 887 static int netvsc_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
 888 {
 889         struct net_device_context *ndc = netdev_priv(dev);
 890
 891         ethtool_cmd_speed_set(cmd, ndc->speed);
 892         cmd->duplex = ndc->duplex;
 893         cmd->port = PORT_OTHER;
 894
 895         return 0;
 896 }
 897
 898 static int netvsc_set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
 899 {
 900         struct net_device_context *ndc = netdev_priv(dev);
 901         u32 speed;
 902
 903         speed = ethtool_cmd_speed(cmd);
 904         if (!ethtool_validate_speed(speed) ||
 905             !ethtool_validate_duplex(cmd->duplex) ||
 906             !netvsc_validate_ethtool_ss_cmd(cmd))
 907                 return -EINVAL;
 908
 909         ndc->speed = speed;
 910         ndc->duplex = cmd->duplex;
 911
 912         return 0;
 913 }
 914
 915 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
 916 {
 917         struct net_device_context *ndevctx = netdev_priv(ndev);
 918         struct netvsc_device *nvdev = ndevctx->nvdev;
 919         struct hv_device *hdev = ndevctx->device_ctx;
 920         struct netvsc_device_info device_info;
 921         int limit = ETH_DATA_LEN;
 922         u32 num_chn;
 923         int ret = 0;
 924
 925         if (ndevctx->start_remove || !nvdev || nvdev->destroy)
 926                 return -ENODEV;
 927
 928         if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
 929                 limit = NETVSC_MTU - ETH_HLEN;
 930
 931         if (mtu < NETVSC_MTU_MIN || mtu > limit)
 932                 return -EINVAL;
 933
 934         ret = netvsc_close(ndev);
 935         if (ret)
 936                 goto out;
 937
 938         num_chn = nvdev->num_chn;
 939
 940         ndevctx->start_remove = true;
 941         rndis_filter_device_remove(hdev);
 942
 943         ndev->mtu = mtu;
 944
 945         memset(&device_info, 0, sizeof(device_info));
 946         device_info.ring_size = ring_size;
 947         device_info.num_chn = num_chn;
 948         device_info.max_num_vrss_chns = max_num_vrss_chns;
 949         rndis_filter_device_add(hdev, &device_info);
 950
 951 out:
 952         netvsc_open(ndev);
 953         ndevctx->start_remove = false;
 954
 955         /* We may have missed link change notifications */
 956         schedule_delayed_work(&ndevctx->dwork, 0);
 957
 958         return ret;
 959 }
 960
 961 static struct rtnl_link_stats64 *netvsc_get_stats64(struct net_device *net,
 962                                                     struct rtnl_link_stats64 *t)
 963 {
 964         struct net_device_context *ndev_ctx = netdev_priv(net);
 965         int cpu;
 966
 967         for_each_possible_cpu(cpu) {
 968                 struct netvsc_stats *tx_stats = per_cpu_ptr(ndev_ctx->tx_stats,
 969                                                             cpu);
 970                 struct netvsc_stats *rx_stats = per_cpu_ptr(ndev_ctx->rx_stats,
 971                                                             cpu);
 972                 u64 tx_packets, tx_bytes, rx_packets, rx_bytes;
 973                 unsigned int start;
 974
 975                 do {
 976                         start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
 977                         tx_packets = tx_stats->packets;
 978                         tx_bytes = tx_stats->bytes;
 979                 } while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
 980
 981                 do {
 982                         start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
 983                         rx_packets = rx_stats->packets;
 984                         rx_bytes = rx_stats->bytes;
 985                 } while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
 986
 987                 t->tx_bytes     += tx_bytes;
 988                 t->tx_packets   += tx_packets;
 989                 t->rx_bytes     += rx_bytes;
 990                 t->rx_packets   += rx_packets;
 991         }
 992
 993         t->tx_dropped   = net->stats.tx_dropped;
 994         t->tx_errors    = net->stats.tx_dropped;
 995
 996         t->rx_dropped   = net->stats.rx_dropped;
 997         t->rx_errors    = net->stats.rx_errors;
 998
 999         return t;
1000 }
1001
1002 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
1003 {
1004         struct sockaddr *addr = p;
1005         char save_adr[ETH_ALEN];
1006         unsigned char save_aatype;
1007         int err;
1008
1009         memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
1010         save_aatype = ndev->addr_assign_type;
1011
1012         err = eth_mac_addr(ndev, p);
1013         if (err != 0)
1014                 return err;
1015
1016         err = rndis_filter_set_device_mac(ndev, addr->sa_data);
1017         if (err != 0) {
1018                 /* roll back to saved MAC */
1019                 memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
1020                 ndev->addr_assign_type = save_aatype;
1021         }
1022
1023         return err;
1024 }
1025
1026 static const struct {
1027         char name[ETH_GSTRING_LEN];
1028         u16 offset;
1029 } netvsc_stats[] = {
1030         { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
1031         { "tx_no_memory",  offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
1032         { "tx_no_space",  offsetof(struct netvsc_ethtool_stats, tx_no_space) },
1033         { "tx_too_big",   offsetof(struct netvsc_ethtool_stats, tx_too_big) },
1034         { "tx_busy",      offsetof(struct netvsc_ethtool_stats, tx_busy) },
1035 };
1036
1037 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
1038 {
1039         switch (string_set) {
1040         case ETH_SS_STATS:
1041                 return ARRAY_SIZE(netvsc_stats);
1042         default:
1043                 return -EINVAL;
1044         }
1045 }
1046
1047 static void netvsc_get_ethtool_stats(struct net_device *dev,
1048                                      struct ethtool_stats *stats, u64 *data)
1049 {
1050         struct net_device_context *ndc = netdev_priv(dev);
1051         const void *nds = &ndc->eth_stats;
1052         int i;
1053
1054         for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++)
1055                 data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1056 }
1057
1058 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1059 {
1060         int i;
1061
1062         switch (stringset) {
1063         case ETH_SS_STATS:
1064                 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++)
1065                         memcpy(data + i * ETH_GSTRING_LEN,
1066                                netvsc_stats[i].name, ETH_GSTRING_LEN);
1067                 break;
1068         }
1069 }
1070
1071 #ifdef CONFIG_NET_POLL_CONTROLLER
1072 static void netvsc_poll_controller(struct net_device *net)
1073 {
1074         /* As netvsc_start_xmit() works synchronous we don't have to
1075          * trigger anything here.
1076          */
1077 }
1078 #endif
1079
1080 static const struct ethtool_ops ethtool_ops = {
1081         .get_drvinfo    = netvsc_get_drvinfo,
1082         .get_link       = ethtool_op_get_link,
1083         .get_ethtool_stats = netvsc_get_ethtool_stats,
1084         .get_sset_count = netvsc_get_sset_count,
1085         .get_strings    = netvsc_get_strings,
1086         .get_channels   = netvsc_get_channels,
1087         .set_channels   = netvsc_set_channels,
1088         .get_ts_info    = ethtool_op_get_ts_info,
1089         .get_settings   = netvsc_get_settings,
1090         .set_settings   = netvsc_set_settings,
1091 };
1092
1093 static const struct net_device_ops device_ops = {
1094         .ndo_open =                     netvsc_open,
1095         .ndo_stop =                     netvsc_close,
1096         .ndo_start_xmit =               netvsc_start_xmit,
1097         .ndo_set_rx_mode =              netvsc_set_multicast_list,
1098         .ndo_change_mtu =               netvsc_change_mtu,
1099         .ndo_validate_addr =            eth_validate_addr,
1100         .ndo_set_mac_address =          netvsc_set_mac_addr,
1101         .ndo_select_queue =             netvsc_select_queue,
1102         .ndo_get_stats64 =              netvsc_get_stats64,
1103 #ifdef CONFIG_NET_POLL_CONTROLLER
1104         .ndo_poll_controller =          netvsc_poll_controller,
1105 #endif
1106 };
1107
1108 /*
1109  * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
1110  * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
1111  * present send GARP packet to network peers with netif_notify_peers().
1112  */
1113 static void netvsc_link_change(struct work_struct *w)
1114 {
1115         struct net_device_context *ndev_ctx =
1116                 container_of(w, struct net_device_context, dwork.work);
1117         struct hv_device *device_obj = ndev_ctx->device_ctx;
1118         struct net_device *net = hv_get_drvdata(device_obj);
1119         struct netvsc_device *net_device;
1120         struct rndis_device *rdev;
1121         struct netvsc_reconfig *event = NULL;
1122         bool notify = false, reschedule = false;
1123         unsigned long flags, next_reconfig, delay;
1124
1125         rtnl_lock();
1126         if (ndev_ctx->start_remove)
1127                 goto out_unlock;
1128
1129         net_device = ndev_ctx->nvdev;
1130         rdev = net_device->extension;
1131
1132         next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
1133         if (time_is_after_jiffies(next_reconfig)) {
1134                 /* link_watch only sends one notification with current state
1135                  * per second, avoid doing reconfig more frequently. Handle
1136                  * wrap around.
1137                  */
1138                 delay = next_reconfig - jiffies;
1139                 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
1140                 schedule_delayed_work(&ndev_ctx->dwork, delay);
1141                 goto out_unlock;
1142         }
1143         ndev_ctx->last_reconfig = jiffies;
1144
1145         spin_lock_irqsave(&ndev_ctx->lock, flags);
1146         if (!list_empty(&ndev_ctx->reconfig_events)) {
1147                 event = list_first_entry(&ndev_ctx->reconfig_events,
1148                                          struct netvsc_reconfig, list);
1149                 list_del(&event->list);
1150                 reschedule = !list_empty(&ndev_ctx->reconfig_events);
1151         }
1152         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1153
1154         if (!event)
1155                 goto out_unlock;
1156
1157         switch (event->event) {
1158                 /* Only the following events are possible due to the check in
1159                  * netvsc_linkstatus_callback()
1160                  */
1161         case RNDIS_STATUS_MEDIA_CONNECT:
1162                 if (rdev->link_state) {
1163                         rdev->link_state = false;
1164                         netif_carrier_on(net);
1165                         netif_tx_wake_all_queues(net);
1166                 } else {
1167                         notify = true;
1168                 }
1169                 kfree(event);
1170                 break;
1171         case RNDIS_STATUS_MEDIA_DISCONNECT:
1172                 if (!rdev->link_state) {
1173                         rdev->link_state = true;
1174                         netif_carrier_off(net);
1175                         netif_tx_stop_all_queues(net);
1176                 }
1177                 kfree(event);
1178                 break;
1179         case RNDIS_STATUS_NETWORK_CHANGE:
1180                 /* Only makes sense if carrier is present */
1181                 if (!rdev->link_state) {
1182                         rdev->link_state = true;
1183                         netif_carrier_off(net);
1184                         netif_tx_stop_all_queues(net);
1185                         event->event = RNDIS_STATUS_MEDIA_CONNECT;
1186                         spin_lock_irqsave(&ndev_ctx->lock, flags);
1187                         list_add(&event->list, &ndev_ctx->reconfig_events);
1188                         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1189                         reschedule = true;
1190                 }
1191                 break;
1192         }
1193
1194         rtnl_unlock();
1195
1196         if (notify)
1197                 netdev_notify_peers(net);
1198
1199         /* link_watch only sends one notification with current state per
1200          * second, handle next reconfig event in 2 seconds.
1201          */
1202         if (reschedule)
1203                 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1204
1205         return;
1206
1207 out_unlock:
1208         rtnl_unlock();
1209 }
1210
1211 static void netvsc_free_netdev(struct net_device *netdev)
1212 {
1213         struct net_device_context *net_device_ctx = netdev_priv(netdev);
1214
1215         free_percpu(net_device_ctx->tx_stats);
1216         free_percpu(net_device_ctx->rx_stats);
1217         free_netdev(netdev);
1218 }
1219
1220 static struct net_device *get_netvsc_bymac(const u8 *mac)
1221 {
1222         struct net_device *dev;
1223
1224         ASSERT_RTNL();
1225
1226         for_each_netdev(&init_net, dev) {
1227                 if (dev->netdev_ops != &device_ops)
1228                         continue;       /* not a netvsc device */
1229
1230                 if (ether_addr_equal(mac, dev->perm_addr))
1231                         return dev;
1232         }
1233
1234         return NULL;
1235 }
1236
1237 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
1238 {
1239         struct net_device *dev;
1240
1241         ASSERT_RTNL();
1242
1243         for_each_netdev(&init_net, dev) {
1244                 struct net_device_context *net_device_ctx;
1245
1246                 if (dev->netdev_ops != &device_ops)
1247                         continue;       /* not a netvsc device */
1248
1249                 net_device_ctx = netdev_priv(dev);
1250                 if (net_device_ctx->nvdev == NULL)
1251                         continue;       /* device is removed */
1252
1253                 if (rtnl_dereference(net_device_ctx->vf_netdev) == vf_netdev)
1254                         return dev;     /* a match */
1255         }
1256
1257         return NULL;
1258 }
1259
1260 static int netvsc_register_vf(struct net_device *vf_netdev)
1261 {
1262         struct net_device *ndev;
1263         struct net_device_context *net_device_ctx;
1264         struct netvsc_device *netvsc_dev;
1265
1266         if (vf_netdev->addr_len != ETH_ALEN)
1267                 return NOTIFY_DONE;
1268
1269         /*
1270          * We will use the MAC address to locate the synthetic interface to
1271          * associate with the VF interface. If we don't find a matching
1272          * synthetic interface, move on.
1273          */
1274         ndev = get_netvsc_bymac(vf_netdev->perm_addr);
1275         if (!ndev)
1276                 return NOTIFY_DONE;
1277
1278         net_device_ctx = netdev_priv(ndev);
1279         netvsc_dev = net_device_ctx->nvdev;
1280         if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
1281                 return NOTIFY_DONE;
1282
1283         netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
1284         /*
1285          * Take a reference on the module.
1286          */
1287         try_module_get(THIS_MODULE);
1288
1289         dev_hold(vf_netdev);
1290         rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
1291         return NOTIFY_OK;
1292 }
1293
1294 static void netvsc_inject_enable(struct net_device_context *net_device_ctx)
1295 {
1296         net_device_ctx->vf_inject = true;
1297 }
1298
1299 static void netvsc_inject_disable(struct net_device_context *net_device_ctx)
1300 {
1301         net_device_ctx->vf_inject = false;
1302
1303         /* Wait for currently active users to drain out. */
1304         while (atomic_read(&net_device_ctx->vf_use_cnt) != 0)
1305                 udelay(50);
1306 }
1307
1308 static int netvsc_vf_up(struct net_device *vf_netdev)
1309 {
1310         struct net_device *ndev;
1311         struct netvsc_device *netvsc_dev;
1312         struct net_device_context *net_device_ctx;
1313
1314         ndev = get_netvsc_byref(vf_netdev);
1315         if (!ndev)
1316                 return NOTIFY_DONE;
1317
1318         net_device_ctx = netdev_priv(ndev);
1319         netvsc_dev = net_device_ctx->nvdev;
1320
1321         netdev_info(ndev, "VF up: %s\n", vf_netdev->name);
1322         netvsc_inject_enable(net_device_ctx);
1323
1324         /*
1325          * Open the device before switching data path.
1326          */
1327         rndis_filter_open(netvsc_dev);
1328
1329         /*
1330          * notify the host to switch the data path.
1331          */
1332         netvsc_switch_datapath(ndev, true);
1333         netdev_info(ndev, "Data path switched to VF: %s\n", vf_netdev->name);
1334
1335         netif_carrier_off(ndev);
1336
1337         /* Now notify peers through VF device. */
1338         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, vf_netdev);
1339
1340         return NOTIFY_OK;
1341 }
1342
1343 static int netvsc_vf_down(struct net_device *vf_netdev)
1344 {
1345         struct net_device *ndev;
1346         struct netvsc_device *netvsc_dev;
1347         struct net_device_context *net_device_ctx;
1348
1349         ndev = get_netvsc_byref(vf_netdev);
1350         if (!ndev)
1351                 return NOTIFY_DONE;
1352
1353         net_device_ctx = netdev_priv(ndev);
1354         netvsc_dev = net_device_ctx->nvdev;
1355
1356         netdev_info(ndev, "VF down: %s\n", vf_netdev->name);
1357         netvsc_inject_disable(net_device_ctx);
1358         netvsc_switch_datapath(ndev, false);
1359         netdev_info(ndev, "Data path switched from VF: %s\n", vf_netdev->name);
1360         rndis_filter_close(netvsc_dev);
1361         netif_carrier_on(ndev);
1362
1363         /* Now notify peers through netvsc device. */
1364         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, ndev);
1365
1366         return NOTIFY_OK;
1367 }
1368
1369 static int netvsc_unregister_vf(struct net_device *vf_netdev)
1370 {
1371         struct net_device *ndev;
1372         struct netvsc_device *netvsc_dev;
1373         struct net_device_context *net_device_ctx;
1374
1375         ndev = get_netvsc_byref(vf_netdev);
1376         if (!ndev)
1377                 return NOTIFY_DONE;
1378
1379         net_device_ctx = netdev_priv(ndev);
1380         netvsc_dev = net_device_ctx->nvdev;
1381
1382         netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
1383         netvsc_inject_disable(net_device_ctx);
1384
1385         RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
1386         dev_put(vf_netdev);
1387         module_put(THIS_MODULE);
1388         return NOTIFY_OK;
1389 }
1390
1391 static int netvsc_probe(struct hv_device *dev,
1392                         const struct hv_vmbus_device_id *dev_id)
1393 {
1394         struct net_device *net = NULL;
1395         struct net_device_context *net_device_ctx;
1396         struct netvsc_device_info device_info;
1397         struct netvsc_device *nvdev;
1398         int ret;
1399
1400         net = alloc_etherdev_mq(sizeof(struct net_device_context),
1401                                 num_online_cpus());
1402         if (!net)
1403                 return -ENOMEM;
1404
1405         netif_carrier_off(net);
1406
1407         netvsc_init_settings(net);
1408
1409         net_device_ctx = netdev_priv(net);
1410         net_device_ctx->device_ctx = dev;
1411         net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
1412         if (netif_msg_probe(net_device_ctx))
1413                 netdev_dbg(net, "netvsc msg_enable: %d\n",
1414                            net_device_ctx->msg_enable);
1415
1416         net_device_ctx->tx_stats = netdev_alloc_pcpu_stats(struct netvsc_stats);
1417         if (!net_device_ctx->tx_stats) {
1418                 free_netdev(net);
1419                 return -ENOMEM;
1420         }
1421         net_device_ctx->rx_stats = netdev_alloc_pcpu_stats(struct netvsc_stats);
1422         if (!net_device_ctx->rx_stats) {
1423                 free_percpu(net_device_ctx->tx_stats);
1424                 free_netdev(net);
1425                 return -ENOMEM;
1426         }
1427
1428         hv_set_drvdata(dev, net);
1429
1430         net_device_ctx->start_remove = false;
1431
1432         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
1433         INIT_WORK(&net_device_ctx->work, do_set_multicast);
1434
1435         spin_lock_init(&net_device_ctx->lock);
1436         INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
1437
1438         atomic_set(&net_device_ctx->vf_use_cnt, 0);
1439
1440         net->netdev_ops = &device_ops;
1441
1442         net->hw_features = NETVSC_HW_FEATURES;
1443         net->features = NETVSC_HW_FEATURES | NETIF_F_HW_VLAN_CTAG_TX;
1444
1445         net->ethtool_ops = &ethtool_ops;
1446         SET_NETDEV_DEV(net, &dev->device);
1447
1448         /* We always need headroom for rndis header */
1449         net->needed_headroom = RNDIS_AND_PPI_SIZE;
1450
1451         /* Notify the netvsc driver of the new device */
1452         memset(&device_info, 0, sizeof(device_info));
1453         device_info.ring_size = ring_size;
1454         device_info.max_num_vrss_chns = max_num_vrss_chns;
1455         ret = rndis_filter_device_add(dev, &device_info);
1456         if (ret != 0) {
1457                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
1458                 netvsc_free_netdev(net);
1459                 hv_set_drvdata(dev, NULL);
1460                 return ret;
1461         }
1462         memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
1463
1464         nvdev = net_device_ctx->nvdev;
1465         netif_set_real_num_tx_queues(net, nvdev->num_chn);
1466         netif_set_real_num_rx_queues(net, nvdev->num_chn);
1467
1468         ret = register_netdev(net);
1469         if (ret != 0) {
1470                 pr_err("Unable to register netdev.\n");
1471                 rndis_filter_device_remove(dev);
1472                 netvsc_free_netdev(net);
1473         }
1474
1475         return ret;
1476 }
1477
1478 static int netvsc_remove(struct hv_device *dev)
1479 {
1480         struct net_device *net;
1481         struct net_device_context *ndev_ctx;
1482         struct netvsc_device *net_device;
1483
1484         net = hv_get_drvdata(dev);
1485
1486         if (net == NULL) {
1487                 dev_err(&dev->device, "No net device to remove\n");
1488                 return 0;
1489         }
1490
1491         ndev_ctx = netdev_priv(net);
1492         net_device = ndev_ctx->nvdev;
1493
1494         /* Avoid racing with netvsc_change_mtu()/netvsc_set_channels()
1495          * removing the device.
1496          */
1497         rtnl_lock();
1498         ndev_ctx->start_remove = true;
1499         rtnl_unlock();
1500
1501         cancel_delayed_work_sync(&ndev_ctx->dwork);
1502         cancel_work_sync(&ndev_ctx->work);
1503
1504         /* Stop outbound asap */
1505         netif_tx_disable(net);
1506
1507         unregister_netdev(net);
1508
1509         /*
1510          * Call to the vsc driver to let it know that the device is being
1511          * removed
1512          */
1513         rndis_filter_device_remove(dev);
1514
1515         hv_set_drvdata(dev, NULL);
1516
1517         netvsc_free_netdev(net);
1518         return 0;
1519 }
1520
1521 static const struct hv_vmbus_device_id id_table[] = {
1522         /* Network guid */
1523         { HV_NIC_GUID, },
1524         { },
1525 };
1526
1527 MODULE_DEVICE_TABLE(vmbus, id_table);
1528
1529 /* The one and only one */
1530 static struct  hv_driver netvsc_drv = {
1531         .name = KBUILD_MODNAME,
1532         .id_table = id_table,
1533         .probe = netvsc_probe,
1534         .remove = netvsc_remove,
1535 };
1536
1537 /*
1538  * On Hyper-V, every VF interface is matched with a corresponding
1539  * synthetic interface. The synthetic interface is presented first
1540  * to the guest. When the corresponding VF instance is registered,
1541  * we will take care of switching the data path.
1542  */
1543 static int netvsc_netdev_event(struct notifier_block *this,
1544                                unsigned long event, void *ptr)
1545 {
1546         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
1547
1548         /* Skip our own events */
1549         if (event_dev->netdev_ops == &device_ops)
1550                 return NOTIFY_DONE;
1551
1552         /* Avoid non-Ethernet type devices */
1553         if (event_dev->type != ARPHRD_ETHER)
1554                 return NOTIFY_DONE;
1555
1556         /* Avoid Vlan dev with same MAC registering as VF */
1557         if (event_dev->priv_flags & IFF_802_1Q_VLAN)
1558                 return NOTIFY_DONE;
1559
1560         /* Avoid Bonding master dev with same MAC registering as VF */
1561         if ((event_dev->priv_flags & IFF_BONDING) &&
1562             (event_dev->flags & IFF_MASTER))
1563                 return NOTIFY_DONE;
1564
1565         switch (event) {
1566         case NETDEV_REGISTER:
1567                 return netvsc_register_vf(event_dev);
1568         case NETDEV_UNREGISTER:
1569                 return netvsc_unregister_vf(event_dev);
1570         case NETDEV_UP:
1571                 return netvsc_vf_up(event_dev);
1572         case NETDEV_DOWN:
1573                 return netvsc_vf_down(event_dev);
1574         default:
1575                 return NOTIFY_DONE;
1576         }
1577 }
1578
1579 static struct notifier_block netvsc_netdev_notifier = {
1580         .notifier_call = netvsc_netdev_event,
1581 };
1582
1583 static void __exit netvsc_drv_exit(void)
1584 {
1585         unregister_netdevice_notifier(&netvsc_netdev_notifier);
1586         vmbus_driver_unregister(&netvsc_drv);
1587 }
1588
1589 static int __init netvsc_drv_init(void)
1590 {
1591         int ret;
1592
1593         if (ring_size < RING_SIZE_MIN) {
1594                 ring_size = RING_SIZE_MIN;
1595                 pr_info("Increased ring_size to %d (min allowed)\n",
1596                         ring_size);
1597         }
1598         ret = vmbus_driver_register(&netvsc_drv);
1599
1600         if (ret)
1601                 return ret;
1602
1603         register_netdevice_notifier(&netvsc_netdev_notifier);
1604         return 0;
1605 }
1606
1607 MODULE_LICENSE("GPL");
1608 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
1609
1610 module_init(netvsc_drv_init);
1611 module_exit(netvsc_drv_exit);