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 sk_buff *skb;
 671         struct sk_buff *vf_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         if (READ_ONCE(net_device_ctx->vf_inject)) {
 680                 atomic_inc(&net_device_ctx->vf_use_cnt);
 681                 if (!READ_ONCE(net_device_ctx->vf_inject)) {
 682                         /*
 683                          * We raced; just move on.
 684                          */
 685                         atomic_dec(&net_device_ctx->vf_use_cnt);
 686                         goto vf_injection_done;
 687                 }
 688
 689                 /*
 690                  * Inject this packet into the VF inerface.
 691                  * On Hyper-V, multicast and brodcast packets
 692                  * are only delivered on the synthetic interface
 693                  * (after subjecting these to policy filters on
 694                  * the host). Deliver these via the VF interface
 695                  * in the guest.
 696                  */
 697                 vf_skb = netvsc_alloc_recv_skb(net_device_ctx->vf_netdev,
 698                                                packet, csum_info, *data,
 699                                                vlan_tci);
 700                 if (vf_skb != NULL) {
 701                         ++net_device_ctx->vf_netdev->stats.rx_packets;
 702                         net_device_ctx->vf_netdev->stats.rx_bytes +=
 703                                 bytes_recvd;
 704                         netif_receive_skb(vf_skb);
 705                 } else {
 706                         ++net->stats.rx_dropped;
 707                         ret = NVSP_STAT_FAIL;
 708                 }
 709                 atomic_dec(&net_device_ctx->vf_use_cnt);
 710                 return ret;
 711         }
 712
 713 vf_injection_done:
 714         rx_stats = this_cpu_ptr(net_device_ctx->rx_stats);
 715
 716         /* Allocate a skb - TODO direct I/O to pages? */
 717         skb = netvsc_alloc_recv_skb(net, packet, csum_info, *data, vlan_tci);
 718         if (unlikely(!skb)) {
 719                 ++net->stats.rx_dropped;
 720                 return NVSP_STAT_FAIL;
 721         }
 722         skb_record_rx_queue(skb, channel->
 723                             offermsg.offer.sub_channel_index);
 724
 725         u64_stats_update_begin(&rx_stats->syncp);
 726         rx_stats->packets++;
 727         rx_stats->bytes += packet->total_data_buflen;
 728         u64_stats_update_end(&rx_stats->syncp);
 729
 730         /*
 731          * Pass the skb back up. Network stack will deallocate the skb when it
 732          * is done.
 733          * TODO - use NAPI?
 734          */
 735         netif_rx(skb);
 736
 737         return 0;
 738 }
 739
 740 static void netvsc_get_drvinfo(struct net_device *net,
 741                                struct ethtool_drvinfo *info)
 742 {
 743         struct net_device_context *net_device_ctx = netdev_priv(net);
 744         struct hv_device *dev = net_device_ctx->device_ctx;
 745
 746         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
 747         strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
 748         strlcpy(info->bus_info, vmbus_dev_name(dev), sizeof(info->bus_info));
 749 }
 750
 751 static void netvsc_get_channels(struct net_device *net,
 752                                 struct ethtool_channels *channel)
 753 {
 754         struct net_device_context *net_device_ctx = netdev_priv(net);
 755         struct netvsc_device *nvdev = net_device_ctx->nvdev;
 756
 757         if (nvdev) {
 758                 channel->max_combined   = nvdev->max_chn;
 759                 channel->combined_count = nvdev->num_chn;
 760         }
 761 }
 762
 763 static int netvsc_set_channels(struct net_device *net,
 764                                struct ethtool_channels *channels)
 765 {
 766         struct net_device_context *net_device_ctx = netdev_priv(net);
 767         struct hv_device *dev = net_device_ctx->device_ctx;
 768         struct netvsc_device *nvdev = net_device_ctx->nvdev;
 769         struct netvsc_device_info device_info;
 770         u32 num_chn;
 771         u32 max_chn;
 772         int ret = 0;
 773         bool recovering = false;
 774
 775         if (net_device_ctx->start_remove || !nvdev || nvdev->destroy)
 776                 return -ENODEV;
 777
 778         num_chn = nvdev->num_chn;
 779         max_chn = min_t(u32, nvdev->max_chn, num_online_cpus());
 780
 781         if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5) {
 782                 pr_info("vRSS unsupported before NVSP Version 5\n");
 783                 return -EINVAL;
 784         }
 785
 786         /* We do not support rx, tx, or other */
 787         if (!channels ||
 788             channels->rx_count ||
 789             channels->tx_count ||
 790             channels->other_count ||
 791             (channels->combined_count < 1))
 792                 return -EINVAL;
 793
 794         if (channels->combined_count > max_chn) {
 795                 pr_info("combined channels too high, using %d\n", max_chn);
 796                 channels->combined_count = max_chn;
 797         }
 798
 799         ret = netvsc_close(net);
 800         if (ret)
 801                 goto out;
 802
 803  do_set:
 804         net_device_ctx->start_remove = true;
 805         rndis_filter_device_remove(dev);
 806
 807         nvdev->num_chn = channels->combined_count;
 808
 809         memset(&device_info, 0, sizeof(device_info));
 810         device_info.num_chn = nvdev->num_chn; /* passed to RNDIS */
 811         device_info.ring_size = ring_size;
 812         device_info.max_num_vrss_chns = max_num_vrss_chns;
 813
 814         ret = rndis_filter_device_add(dev, &device_info);
 815         if (ret) {
 816                 if (recovering) {
 817                         netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
 818                         return ret;
 819                 }
 820                 goto recover;
 821         }
 822
 823         nvdev = net_device_ctx->nvdev;
 824
 825         ret = netif_set_real_num_tx_queues(net, nvdev->num_chn);
 826         if (ret) {
 827                 if (recovering) {
 828                         netdev_err(net, "could not set tx queue count (ret %d)\n", ret);
 829                         return ret;
 830                 }
 831                 goto recover;
 832         }
 833
 834         ret = netif_set_real_num_rx_queues(net, nvdev->num_chn);
 835         if (ret) {
 836                 if (recovering) {
 837                         netdev_err(net, "could not set rx queue count (ret %d)\n", ret);
 838                         return ret;
 839                 }
 840                 goto recover;
 841         }
 842
 843  out:
 844         netvsc_open(net);
 845         net_device_ctx->start_remove = false;
 846         /* We may have missed link change notifications */
 847         schedule_delayed_work(&net_device_ctx->dwork, 0);
 848
 849         return ret;
 850
 851  recover:
 852         /* If the above failed, we attempt to recover through the same
 853          * process but with the original number of channels.
 854          */
 855         netdev_err(net, "could not set channels, recovering\n");
 856         recovering = true;
 857         channels->combined_count = num_chn;
 858         goto do_set;
 859 }
 860
 861 static bool netvsc_validate_ethtool_ss_cmd(const struct ethtool_cmd *cmd)
 862 {
 863         struct ethtool_cmd diff1 = *cmd;
 864         struct ethtool_cmd diff2 = {};
 865
 866         ethtool_cmd_speed_set(&diff1, 0);
 867         diff1.duplex = 0;
 868         /* advertising and cmd are usually set */
 869         diff1.advertising = 0;
 870         diff1.cmd = 0;
 871         /* We set port to PORT_OTHER */
 872         diff2.port = PORT_OTHER;
 873
 874         return !memcmp(&diff1, &diff2, sizeof(diff1));
 875 }
 876
 877 static void netvsc_init_settings(struct net_device *dev)
 878 {
 879         struct net_device_context *ndc = netdev_priv(dev);
 880
 881         ndc->speed = SPEED_UNKNOWN;
 882         ndc->duplex = DUPLEX_UNKNOWN;
 883 }
 884
 885 static int netvsc_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
 886 {
 887         struct net_device_context *ndc = netdev_priv(dev);
 888
 889         ethtool_cmd_speed_set(cmd, ndc->speed);
 890         cmd->duplex = ndc->duplex;
 891         cmd->port = PORT_OTHER;
 892
 893         return 0;
 894 }
 895
 896 static int netvsc_set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
 897 {
 898         struct net_device_context *ndc = netdev_priv(dev);
 899         u32 speed;
 900
 901         speed = ethtool_cmd_speed(cmd);
 902         if (!ethtool_validate_speed(speed) ||
 903             !ethtool_validate_duplex(cmd->duplex) ||
 904             !netvsc_validate_ethtool_ss_cmd(cmd))
 905                 return -EINVAL;
 906
 907         ndc->speed = speed;
 908         ndc->duplex = cmd->duplex;
 909
 910         return 0;
 911 }
 912
 913 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
 914 {
 915         struct net_device_context *ndevctx = netdev_priv(ndev);
 916         struct netvsc_device *nvdev = ndevctx->nvdev;
 917         struct hv_device *hdev = ndevctx->device_ctx;
 918         struct netvsc_device_info device_info;
 919         int limit = ETH_DATA_LEN;
 920         u32 num_chn;
 921         int ret = 0;
 922
 923         if (ndevctx->start_remove || !nvdev || nvdev->destroy)
 924                 return -ENODEV;
 925
 926         if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
 927                 limit = NETVSC_MTU - ETH_HLEN;
 928
 929         if (mtu < NETVSC_MTU_MIN || mtu > limit)
 930                 return -EINVAL;
 931
 932         ret = netvsc_close(ndev);
 933         if (ret)
 934                 goto out;
 935
 936         num_chn = nvdev->num_chn;
 937
 938         ndevctx->start_remove = true;
 939         rndis_filter_device_remove(hdev);
 940
 941         ndev->mtu = mtu;
 942
 943         memset(&device_info, 0, sizeof(device_info));
 944         device_info.ring_size = ring_size;
 945         device_info.num_chn = num_chn;
 946         device_info.max_num_vrss_chns = max_num_vrss_chns;
 947         rndis_filter_device_add(hdev, &device_info);
 948
 949 out:
 950         netvsc_open(ndev);
 951         ndevctx->start_remove = false;
 952
 953         /* We may have missed link change notifications */
 954         schedule_delayed_work(&ndevctx->dwork, 0);
 955
 956         return ret;
 957 }
 958
 959 static struct rtnl_link_stats64 *netvsc_get_stats64(struct net_device *net,
 960                                                     struct rtnl_link_stats64 *t)
 961 {
 962         struct net_device_context *ndev_ctx = netdev_priv(net);
 963         int cpu;
 964
 965         for_each_possible_cpu(cpu) {
 966                 struct netvsc_stats *tx_stats = per_cpu_ptr(ndev_ctx->tx_stats,
 967                                                             cpu);
 968                 struct netvsc_stats *rx_stats = per_cpu_ptr(ndev_ctx->rx_stats,
 969                                                             cpu);
 970                 u64 tx_packets, tx_bytes, rx_packets, rx_bytes;
 971                 unsigned int start;
 972
 973                 do {
 974                         start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
 975                         tx_packets = tx_stats->packets;
 976                         tx_bytes = tx_stats->bytes;
 977                 } while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
 978
 979                 do {
 980                         start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
 981                         rx_packets = rx_stats->packets;
 982                         rx_bytes = rx_stats->bytes;
 983                 } while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
 984
 985                 t->tx_bytes     += tx_bytes;
 986                 t->tx_packets   += tx_packets;
 987                 t->rx_bytes     += rx_bytes;
 988                 t->rx_packets   += rx_packets;
 989         }
 990
 991         t->tx_dropped   = net->stats.tx_dropped;
 992         t->tx_errors    = net->stats.tx_dropped;
 993
 994         t->rx_dropped   = net->stats.rx_dropped;
 995         t->rx_errors    = net->stats.rx_errors;
 996
 997         return t;
 998 }
 999
1000 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
1001 {
1002         struct sockaddr *addr = p;
1003         char save_adr[ETH_ALEN];
1004         unsigned char save_aatype;
1005         int err;
1006
1007         memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
1008         save_aatype = ndev->addr_assign_type;
1009
1010         err = eth_mac_addr(ndev, p);
1011         if (err != 0)
1012                 return err;
1013
1014         err = rndis_filter_set_device_mac(ndev, addr->sa_data);
1015         if (err != 0) {
1016                 /* roll back to saved MAC */
1017                 memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
1018                 ndev->addr_assign_type = save_aatype;
1019         }
1020
1021         return err;
1022 }
1023
1024 static const struct {
1025         char name[ETH_GSTRING_LEN];
1026         u16 offset;
1027 } netvsc_stats[] = {
1028         { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
1029         { "tx_no_memory",  offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
1030         { "tx_no_space",  offsetof(struct netvsc_ethtool_stats, tx_no_space) },
1031         { "tx_too_big",   offsetof(struct netvsc_ethtool_stats, tx_too_big) },
1032         { "tx_busy",      offsetof(struct netvsc_ethtool_stats, tx_busy) },
1033 };
1034
1035 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
1036 {
1037         switch (string_set) {
1038         case ETH_SS_STATS:
1039                 return ARRAY_SIZE(netvsc_stats);
1040         default:
1041                 return -EINVAL;
1042         }
1043 }
1044
1045 static void netvsc_get_ethtool_stats(struct net_device *dev,
1046                                      struct ethtool_stats *stats, u64 *data)
1047 {
1048         struct net_device_context *ndc = netdev_priv(dev);
1049         const void *nds = &ndc->eth_stats;
1050         int i;
1051
1052         for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++)
1053                 data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1054 }
1055
1056 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1057 {
1058         int i;
1059
1060         switch (stringset) {
1061         case ETH_SS_STATS:
1062                 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++)
1063                         memcpy(data + i * ETH_GSTRING_LEN,
1064                                netvsc_stats[i].name, ETH_GSTRING_LEN);
1065                 break;
1066         }
1067 }
1068
1069 #ifdef CONFIG_NET_POLL_CONTROLLER
1070 static void netvsc_poll_controller(struct net_device *net)
1071 {
1072         /* As netvsc_start_xmit() works synchronous we don't have to
1073          * trigger anything here.
1074          */
1075 }
1076 #endif
1077
1078 static const struct ethtool_ops ethtool_ops = {
1079         .get_drvinfo    = netvsc_get_drvinfo,
1080         .get_link       = ethtool_op_get_link,
1081         .get_ethtool_stats = netvsc_get_ethtool_stats,
1082         .get_sset_count = netvsc_get_sset_count,
1083         .get_strings    = netvsc_get_strings,
1084         .get_channels   = netvsc_get_channels,
1085         .set_channels   = netvsc_set_channels,
1086         .get_ts_info    = ethtool_op_get_ts_info,
1087         .get_settings   = netvsc_get_settings,
1088         .set_settings   = netvsc_set_settings,
1089 };
1090
1091 static const struct net_device_ops device_ops = {
1092         .ndo_open =                     netvsc_open,
1093         .ndo_stop =                     netvsc_close,
1094         .ndo_start_xmit =               netvsc_start_xmit,
1095         .ndo_set_rx_mode =              netvsc_set_multicast_list,
1096         .ndo_change_mtu =               netvsc_change_mtu,
1097         .ndo_validate_addr =            eth_validate_addr,
1098         .ndo_set_mac_address =          netvsc_set_mac_addr,
1099         .ndo_select_queue =             netvsc_select_queue,
1100         .ndo_get_stats64 =              netvsc_get_stats64,
1101 #ifdef CONFIG_NET_POLL_CONTROLLER
1102         .ndo_poll_controller =          netvsc_poll_controller,
1103 #endif
1104 };
1105
1106 /*
1107  * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
1108  * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
1109  * present send GARP packet to network peers with netif_notify_peers().
1110  */
1111 static void netvsc_link_change(struct work_struct *w)
1112 {
1113         struct net_device_context *ndev_ctx =
1114                 container_of(w, struct net_device_context, dwork.work);
1115         struct hv_device *device_obj = ndev_ctx->device_ctx;
1116         struct net_device *net = hv_get_drvdata(device_obj);
1117         struct netvsc_device *net_device;
1118         struct rndis_device *rdev;
1119         struct netvsc_reconfig *event = NULL;
1120         bool notify = false, reschedule = false;
1121         unsigned long flags, next_reconfig, delay;
1122
1123         rtnl_lock();
1124         if (ndev_ctx->start_remove)
1125                 goto out_unlock;
1126
1127         net_device = ndev_ctx->nvdev;
1128         rdev = net_device->extension;
1129
1130         next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
1131         if (time_is_after_jiffies(next_reconfig)) {
1132                 /* link_watch only sends one notification with current state
1133                  * per second, avoid doing reconfig more frequently. Handle
1134                  * wrap around.
1135                  */
1136                 delay = next_reconfig - jiffies;
1137                 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
1138                 schedule_delayed_work(&ndev_ctx->dwork, delay);
1139                 goto out_unlock;
1140         }
1141         ndev_ctx->last_reconfig = jiffies;
1142
1143         spin_lock_irqsave(&ndev_ctx->lock, flags);
1144         if (!list_empty(&ndev_ctx->reconfig_events)) {
1145                 event = list_first_entry(&ndev_ctx->reconfig_events,
1146                                          struct netvsc_reconfig, list);
1147                 list_del(&event->list);
1148                 reschedule = !list_empty(&ndev_ctx->reconfig_events);
1149         }
1150         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1151
1152         if (!event)
1153                 goto out_unlock;
1154
1155         switch (event->event) {
1156                 /* Only the following events are possible due to the check in
1157                  * netvsc_linkstatus_callback()
1158                  */
1159         case RNDIS_STATUS_MEDIA_CONNECT:
1160                 if (rdev->link_state) {
1161                         rdev->link_state = false;
1162                         netif_carrier_on(net);
1163                         netif_tx_wake_all_queues(net);
1164                 } else {
1165                         notify = true;
1166                 }
1167                 kfree(event);
1168                 break;
1169         case RNDIS_STATUS_MEDIA_DISCONNECT:
1170                 if (!rdev->link_state) {
1171                         rdev->link_state = true;
1172                         netif_carrier_off(net);
1173                         netif_tx_stop_all_queues(net);
1174                 }
1175                 kfree(event);
1176                 break;
1177         case RNDIS_STATUS_NETWORK_CHANGE:
1178                 /* Only makes sense if carrier is present */
1179                 if (!rdev->link_state) {
1180                         rdev->link_state = true;
1181                         netif_carrier_off(net);
1182                         netif_tx_stop_all_queues(net);
1183                         event->event = RNDIS_STATUS_MEDIA_CONNECT;
1184                         spin_lock_irqsave(&ndev_ctx->lock, flags);
1185                         list_add(&event->list, &ndev_ctx->reconfig_events);
1186                         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1187                         reschedule = true;
1188                 }
1189                 break;
1190         }
1191
1192         rtnl_unlock();
1193
1194         if (notify)
1195                 netdev_notify_peers(net);
1196
1197         /* link_watch only sends one notification with current state per
1198          * second, handle next reconfig event in 2 seconds.
1199          */
1200         if (reschedule)
1201                 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1202
1203         return;
1204
1205 out_unlock:
1206         rtnl_unlock();
1207 }
1208
1209 static void netvsc_free_netdev(struct net_device *netdev)
1210 {
1211         struct net_device_context *net_device_ctx = netdev_priv(netdev);
1212
1213         free_percpu(net_device_ctx->tx_stats);
1214         free_percpu(net_device_ctx->rx_stats);
1215         free_netdev(netdev);
1216 }
1217
1218 static struct net_device *get_netvsc_net_device(char *mac)
1219 {
1220         struct net_device *dev, *found = NULL;
1221
1222         ASSERT_RTNL();
1223
1224         for_each_netdev(&init_net, dev) {
1225                 if (memcmp(dev->dev_addr, mac, ETH_ALEN) == 0) {
1226                         if (dev->netdev_ops != &device_ops)
1227                                 continue;
1228                         found = dev;
1229                         break;
1230                 }
1231         }
1232
1233         return found;
1234 }
1235
1236 static int netvsc_register_vf(struct net_device *vf_netdev)
1237 {
1238         struct net_device *ndev;
1239         struct net_device_context *net_device_ctx;
1240         struct netvsc_device *netvsc_dev;
1241         const struct ethtool_ops *eth_ops = vf_netdev->ethtool_ops;
1242
1243         if (eth_ops == NULL || eth_ops == &ethtool_ops)
1244                 return NOTIFY_DONE;
1245
1246         /*
1247          * We will use the MAC address to locate the synthetic interface to
1248          * associate with the VF interface. If we don't find a matching
1249          * synthetic interface, move on.
1250          */
1251         ndev = get_netvsc_net_device(vf_netdev->dev_addr);
1252         if (!ndev)
1253                 return NOTIFY_DONE;
1254
1255         net_device_ctx = netdev_priv(ndev);
1256         netvsc_dev = net_device_ctx->nvdev;
1257         if (!netvsc_dev || net_device_ctx->vf_netdev)
1258                 return NOTIFY_DONE;
1259
1260         netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
1261         /*
1262          * Take a reference on the module.
1263          */
1264         try_module_get(THIS_MODULE);
1265
1266         dev_hold(vf_netdev);
1267         net_device_ctx->vf_netdev = vf_netdev;
1268         return NOTIFY_OK;
1269 }
1270
1271 static void netvsc_inject_enable(struct net_device_context *net_device_ctx)
1272 {
1273         net_device_ctx->vf_inject = true;
1274 }
1275
1276 static void netvsc_inject_disable(struct net_device_context *net_device_ctx)
1277 {
1278         net_device_ctx->vf_inject = false;
1279
1280         /* Wait for currently active users to drain out. */
1281         while (atomic_read(&net_device_ctx->vf_use_cnt) != 0)
1282                 udelay(50);
1283 }
1284
1285 static int netvsc_vf_up(struct net_device *vf_netdev)
1286 {
1287         struct net_device *ndev;
1288         struct netvsc_device *netvsc_dev;
1289         const struct ethtool_ops *eth_ops = vf_netdev->ethtool_ops;
1290         struct net_device_context *net_device_ctx;
1291
1292         if (eth_ops == &ethtool_ops)
1293                 return NOTIFY_DONE;
1294
1295         ndev = get_netvsc_net_device(vf_netdev->dev_addr);
1296         if (!ndev)
1297                 return NOTIFY_DONE;
1298
1299         net_device_ctx = netdev_priv(ndev);
1300         netvsc_dev = net_device_ctx->nvdev;
1301
1302         if (!netvsc_dev || !net_device_ctx->vf_netdev)
1303                 return NOTIFY_DONE;
1304
1305         netdev_info(ndev, "VF up: %s\n", vf_netdev->name);
1306         netvsc_inject_enable(net_device_ctx);
1307
1308         /*
1309          * Open the device before switching data path.
1310          */
1311         rndis_filter_open(netvsc_dev);
1312
1313         /*
1314          * notify the host to switch the data path.
1315          */
1316         netvsc_switch_datapath(ndev, true);
1317         netdev_info(ndev, "Data path switched to VF: %s\n", vf_netdev->name);
1318
1319         netif_carrier_off(ndev);
1320
1321         /* Now notify peers through VF device. */
1322         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, vf_netdev);
1323
1324         return NOTIFY_OK;
1325 }
1326
1327 static int netvsc_vf_down(struct net_device *vf_netdev)
1328 {
1329         struct net_device *ndev;
1330         struct netvsc_device *netvsc_dev;
1331         struct net_device_context *net_device_ctx;
1332         const struct ethtool_ops *eth_ops = vf_netdev->ethtool_ops;
1333
1334         if (eth_ops == &ethtool_ops)
1335                 return NOTIFY_DONE;
1336
1337         ndev = get_netvsc_net_device(vf_netdev->dev_addr);
1338         if (!ndev)
1339                 return NOTIFY_DONE;
1340
1341         net_device_ctx = netdev_priv(ndev);
1342         netvsc_dev = net_device_ctx->nvdev;
1343
1344         if (!netvsc_dev || !net_device_ctx->vf_netdev)
1345                 return NOTIFY_DONE;
1346
1347         netdev_info(ndev, "VF down: %s\n", vf_netdev->name);
1348         netvsc_inject_disable(net_device_ctx);
1349         netvsc_switch_datapath(ndev, false);
1350         netdev_info(ndev, "Data path switched from VF: %s\n", vf_netdev->name);
1351         rndis_filter_close(netvsc_dev);
1352         netif_carrier_on(ndev);
1353
1354         /* Now notify peers through netvsc device. */
1355         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, ndev);
1356
1357         return NOTIFY_OK;
1358 }
1359
1360 static int netvsc_unregister_vf(struct net_device *vf_netdev)
1361 {
1362         struct net_device *ndev;
1363         struct netvsc_device *netvsc_dev;
1364         const struct ethtool_ops *eth_ops = vf_netdev->ethtool_ops;
1365         struct net_device_context *net_device_ctx;
1366
1367         if (eth_ops == &ethtool_ops)
1368                 return NOTIFY_DONE;
1369
1370         ndev = get_netvsc_net_device(vf_netdev->dev_addr);
1371         if (!ndev)
1372                 return NOTIFY_DONE;
1373
1374         net_device_ctx = netdev_priv(ndev);
1375         netvsc_dev = net_device_ctx->nvdev;
1376         if (!netvsc_dev || !net_device_ctx->vf_netdev)
1377                 return NOTIFY_DONE;
1378         netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
1379         netvsc_inject_disable(net_device_ctx);
1380         net_device_ctx->vf_netdev = NULL;
1381         dev_put(vf_netdev);
1382         module_put(THIS_MODULE);
1383         return NOTIFY_OK;
1384 }
1385
1386 static int netvsc_probe(struct hv_device *dev,
1387                         const struct hv_vmbus_device_id *dev_id)
1388 {
1389         struct net_device *net = NULL;
1390         struct net_device_context *net_device_ctx;
1391         struct netvsc_device_info device_info;
1392         struct netvsc_device *nvdev;
1393         int ret;
1394
1395         net = alloc_etherdev_mq(sizeof(struct net_device_context),
1396                                 num_online_cpus());
1397         if (!net)
1398                 return -ENOMEM;
1399
1400         netif_carrier_off(net);
1401
1402         netvsc_init_settings(net);
1403
1404         net_device_ctx = netdev_priv(net);
1405         net_device_ctx->device_ctx = dev;
1406         net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
1407         if (netif_msg_probe(net_device_ctx))
1408                 netdev_dbg(net, "netvsc msg_enable: %d\n",
1409                            net_device_ctx->msg_enable);
1410
1411         net_device_ctx->tx_stats = netdev_alloc_pcpu_stats(struct netvsc_stats);
1412         if (!net_device_ctx->tx_stats) {
1413                 free_netdev(net);
1414                 return -ENOMEM;
1415         }
1416         net_device_ctx->rx_stats = netdev_alloc_pcpu_stats(struct netvsc_stats);
1417         if (!net_device_ctx->rx_stats) {
1418                 free_percpu(net_device_ctx->tx_stats);
1419                 free_netdev(net);
1420                 return -ENOMEM;
1421         }
1422
1423         hv_set_drvdata(dev, net);
1424
1425         net_device_ctx->start_remove = false;
1426
1427         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
1428         INIT_WORK(&net_device_ctx->work, do_set_multicast);
1429
1430         spin_lock_init(&net_device_ctx->lock);
1431         INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
1432
1433         atomic_set(&net_device_ctx->vf_use_cnt, 0);
1434         net_device_ctx->vf_netdev = NULL;
1435         net_device_ctx->vf_inject = false;
1436
1437         net->netdev_ops = &device_ops;
1438
1439         net->hw_features = NETVSC_HW_FEATURES;
1440         net->features = NETVSC_HW_FEATURES | NETIF_F_HW_VLAN_CTAG_TX;
1441
1442         net->ethtool_ops = &ethtool_ops;
1443         SET_NETDEV_DEV(net, &dev->device);
1444
1445         /* We always need headroom for rndis header */
1446         net->needed_headroom = RNDIS_AND_PPI_SIZE;
1447
1448         /* Notify the netvsc driver of the new device */
1449         memset(&device_info, 0, sizeof(device_info));
1450         device_info.ring_size = ring_size;
1451         device_info.max_num_vrss_chns = max_num_vrss_chns;
1452         ret = rndis_filter_device_add(dev, &device_info);
1453         if (ret != 0) {
1454                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
1455                 netvsc_free_netdev(net);
1456                 hv_set_drvdata(dev, NULL);
1457                 return ret;
1458         }
1459         memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
1460
1461         nvdev = net_device_ctx->nvdev;
1462         netif_set_real_num_tx_queues(net, nvdev->num_chn);
1463         netif_set_real_num_rx_queues(net, nvdev->num_chn);
1464
1465         ret = register_netdev(net);
1466         if (ret != 0) {
1467                 pr_err("Unable to register netdev.\n");
1468                 rndis_filter_device_remove(dev);
1469                 netvsc_free_netdev(net);
1470         }
1471
1472         return ret;
1473 }
1474
1475 static int netvsc_remove(struct hv_device *dev)
1476 {
1477         struct net_device *net;
1478         struct net_device_context *ndev_ctx;
1479         struct netvsc_device *net_device;
1480
1481         net = hv_get_drvdata(dev);
1482
1483         if (net == NULL) {
1484                 dev_err(&dev->device, "No net device to remove\n");
1485                 return 0;
1486         }
1487
1488         ndev_ctx = netdev_priv(net);
1489         net_device = ndev_ctx->nvdev;
1490
1491         /* Avoid racing with netvsc_change_mtu()/netvsc_set_channels()
1492          * removing the device.
1493          */
1494         rtnl_lock();
1495         ndev_ctx->start_remove = true;
1496         rtnl_unlock();
1497
1498         cancel_delayed_work_sync(&ndev_ctx->dwork);
1499         cancel_work_sync(&ndev_ctx->work);
1500
1501         /* Stop outbound asap */
1502         netif_tx_disable(net);
1503
1504         unregister_netdev(net);
1505
1506         /*
1507          * Call to the vsc driver to let it know that the device is being
1508          * removed
1509          */
1510         rndis_filter_device_remove(dev);
1511
1512         hv_set_drvdata(dev, NULL);
1513
1514         netvsc_free_netdev(net);
1515         return 0;
1516 }
1517
1518 static const struct hv_vmbus_device_id id_table[] = {
1519         /* Network guid */
1520         { HV_NIC_GUID, },
1521         { },
1522 };
1523
1524 MODULE_DEVICE_TABLE(vmbus, id_table);
1525
1526 /* The one and only one */
1527 static struct  hv_driver netvsc_drv = {
1528         .name = KBUILD_MODNAME,
1529         .id_table = id_table,
1530         .probe = netvsc_probe,
1531         .remove = netvsc_remove,
1532 };
1533
1534 /*
1535  * On Hyper-V, every VF interface is matched with a corresponding
1536  * synthetic interface. The synthetic interface is presented first
1537  * to the guest. When the corresponding VF instance is registered,
1538  * we will take care of switching the data path.
1539  */
1540 static int netvsc_netdev_event(struct notifier_block *this,
1541                                unsigned long event, void *ptr)
1542 {
1543         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
1544
1545         /* Avoid Vlan dev with same MAC registering as VF */
1546         if (event_dev->priv_flags & IFF_802_1Q_VLAN)
1547                 return NOTIFY_DONE;
1548
1549         /* Avoid Bonding master dev with same MAC registering as VF */
1550         if (event_dev->priv_flags & IFF_BONDING &&
1551             event_dev->flags & IFF_MASTER)
1552                 return NOTIFY_DONE;
1553
1554         switch (event) {
1555         case NETDEV_REGISTER:
1556                 return netvsc_register_vf(event_dev);
1557         case NETDEV_UNREGISTER:
1558                 return netvsc_unregister_vf(event_dev);
1559         case NETDEV_UP:
1560                 return netvsc_vf_up(event_dev);
1561         case NETDEV_DOWN:
1562                 return netvsc_vf_down(event_dev);
1563         default:
1564                 return NOTIFY_DONE;
1565         }
1566 }
1567
1568 static struct notifier_block netvsc_netdev_notifier = {
1569         .notifier_call = netvsc_netdev_event,
1570 };
1571
1572 static void __exit netvsc_drv_exit(void)
1573 {
1574         unregister_netdevice_notifier(&netvsc_netdev_notifier);
1575         vmbus_driver_unregister(&netvsc_drv);
1576 }
1577
1578 static int __init netvsc_drv_init(void)
1579 {
1580         int ret;
1581
1582         if (ring_size < RING_SIZE_MIN) {
1583                 ring_size = RING_SIZE_MIN;
1584                 pr_info("Increased ring_size to %d (min allowed)\n",
1585                         ring_size);
1586         }
1587         ret = vmbus_driver_register(&netvsc_drv);
1588
1589         if (ret)
1590                 return ret;
1591
1592         register_netdevice_notifier(&netvsc_netdev_notifier);
1593         return 0;
1594 }
1595
1596 MODULE_LICENSE("GPL");
1597 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
1598
1599 module_init(netvsc_drv_init);
1600 module_exit(netvsc_drv_exit);