drivers/net/hyperv/netvsc_drv.c

   1 /*
   2  * Copyright (c) 2009, Microsoft Corporation.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms and conditions of the GNU General Public License,
   6  * version 2, as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope it will be useful, but WITHOUT
   9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  11  * more details.
  12  *
  13  * You should have received a copy of the GNU General Public License along with
  14  * this program; if not, see <http://www.gnu.org/licenses/>.
  15  *
  16  * Authors:
  17  *   Haiyang Zhang <haiyangz@microsoft.com>
  18  *   Hank Janssen  <hjanssen@microsoft.com>
  19  */
  20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  21
  22 #include <linux/init.h>
  23 #include <linux/atomic.h>
  24 #include <linux/module.h>
  25 #include <linux/highmem.h>
  26 #include <linux/device.h>
  27 #include <linux/io.h>
  28 #include <linux/delay.h>
  29 #include <linux/netdevice.h>
  30 #include <linux/inetdevice.h>
  31 #include <linux/etherdevice.h>
  32 #include <linux/skbuff.h>
  33 #include <linux/if_vlan.h>
  34 #include <linux/in.h>
  35 #include <linux/slab.h>
  36 #include <net/arp.h>
  37 #include <net/route.h>
  38 #include <net/sock.h>
  39 #include <net/pkt_sched.h>
  40
  41 #include "hyperv_net.h"
  42
  43 #define RING_SIZE_MIN 64
  44 #define LINKCHANGE_INT (2 * HZ)
  45
  46 static int ring_size = 128;
  47 module_param(ring_size, int, S_IRUGO);
  48 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
  49
  50 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
  51                                 NETIF_MSG_LINK | NETIF_MSG_IFUP |
  52                                 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
  53                                 NETIF_MSG_TX_ERR;
  54
  55 static int debug = -1;
  56 module_param(debug, int, S_IRUGO);
  57 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  58
  59 static void do_set_multicast(struct work_struct *w)
  60 {
  61         struct net_device_context *ndevctx =
  62                 container_of(w, struct net_device_context, work);
  63         struct hv_device *device_obj = ndevctx->device_ctx;
  64         struct net_device *ndev = hv_get_drvdata(device_obj);
  65         struct netvsc_device *nvdev = rcu_dereference(ndevctx->nvdev);
  66         struct rndis_device *rdev;
  67
  68         if (!nvdev)
  69                 return;
  70
  71         rdev = nvdev->extension;
  72         if (rdev == NULL)
  73                 return;
  74
  75         if (ndev->flags & IFF_PROMISC)
  76                 rndis_filter_set_packet_filter(rdev,
  77                         NDIS_PACKET_TYPE_PROMISCUOUS);
  78         else
  79                 rndis_filter_set_packet_filter(rdev,
  80                         NDIS_PACKET_TYPE_BROADCAST |
  81                         NDIS_PACKET_TYPE_ALL_MULTICAST |
  82                         NDIS_PACKET_TYPE_DIRECTED);
  83 }
  84
  85 static void netvsc_set_multicast_list(struct net_device *net)
  86 {
  87         struct net_device_context *net_device_ctx = netdev_priv(net);
  88
  89         schedule_work(&net_device_ctx->work);
  90 }
  91
  92 static int netvsc_open(struct net_device *net)
  93 {
  94         struct netvsc_device *nvdev = net_device_to_netvsc_device(net);
  95         struct rndis_device *rdev;
  96         int ret = 0;
  97
  98         netif_carrier_off(net);
  99
 100         /* Open up the device */
 101         ret = rndis_filter_open(nvdev);
 102         if (ret != 0) {
 103                 netdev_err(net, "unable to open device (ret %d).\n", ret);
 104                 return ret;
 105         }
 106
 107         netif_tx_wake_all_queues(net);
 108
 109         rdev = nvdev->extension;
 110         if (!rdev->link_state)
 111                 netif_carrier_on(net);
 112
 113         return ret;
 114 }
 115
 116 static int netvsc_close(struct net_device *net)
 117 {
 118         struct net_device_context *net_device_ctx = netdev_priv(net);
 119         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
 120         int ret;
 121         u32 aread, awrite, i, msec = 10, retry = 0, retry_max = 20;
 122         struct vmbus_channel *chn;
 123
 124         netif_tx_disable(net);
 125
 126         /* Make sure netvsc_set_multicast_list doesn't re-enable filter! */
 127         cancel_work_sync(&net_device_ctx->work);
 128         ret = rndis_filter_close(nvdev);
 129         if (ret != 0) {
 130                 netdev_err(net, "unable to close device (ret %d).\n", ret);
 131                 return ret;
 132         }
 133
 134         /* Ensure pending bytes in ring are read */
 135         while (true) {
 136                 aread = 0;
 137                 for (i = 0; i < nvdev->num_chn; i++) {
 138                         chn = nvdev->chan_table[i].channel;
 139                         if (!chn)
 140                                 continue;
 141
 142                         hv_get_ringbuffer_availbytes(&chn->inbound, &aread,
 143                                                      &awrite);
 144
 145                         if (aread)
 146                                 break;
 147
 148                         hv_get_ringbuffer_availbytes(&chn->outbound, &aread,
 149                                                      &awrite);
 150
 151                         if (aread)
 152                                 break;
 153                 }
 154
 155                 retry++;
 156                 if (retry > retry_max || aread == 0)
 157                         break;
 158
 159                 msleep(msec);
 160
 161                 if (msec < 1000)
 162                         msec *= 2;
 163         }
 164
 165         if (aread) {
 166                 netdev_err(net, "Ring buffer not empty after closing rndis\n");
 167                 ret = -ETIMEDOUT;
 168         }
 169
 170         return ret;
 171 }
 172
 173 static void *init_ppi_data(struct rndis_message *msg, u32 ppi_size,
 174                                 int pkt_type)
 175 {
 176         struct rndis_packet *rndis_pkt;
 177         struct rndis_per_packet_info *ppi;
 178
 179         rndis_pkt = &msg->msg.pkt;
 180         rndis_pkt->data_offset += ppi_size;
 181
 182         ppi = (struct rndis_per_packet_info *)((void *)rndis_pkt +
 183                 rndis_pkt->per_pkt_info_offset + rndis_pkt->per_pkt_info_len);
 184
 185         ppi->size = ppi_size;
 186         ppi->type = pkt_type;
 187         ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
 188
 189         rndis_pkt->per_pkt_info_len += ppi_size;
 190
 191         return ppi;
 192 }
 193
 194 /* Azure hosts don't support non-TCP port numbers in hashing yet. We compute
 195  * hash for non-TCP traffic with only IP numbers.
 196  */
 197 static inline u32 netvsc_get_hash(struct sk_buff *skb, struct sock *sk)
 198 {
 199         struct flow_keys flow;
 200         u32 hash;
 201         static u32 hashrnd __read_mostly;
 202
 203         net_get_random_once(&hashrnd, sizeof(hashrnd));
 204
 205         if (!skb_flow_dissect_flow_keys(skb, &flow, 0))
 206                 return 0;
 207
 208         if (flow.basic.ip_proto == IPPROTO_TCP) {
 209                 return skb_get_hash(skb);
 210         } else {
 211                 if (flow.basic.n_proto == htons(ETH_P_IP))
 212                         hash = jhash2((u32 *)&flow.addrs.v4addrs, 2, hashrnd);
 213                 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
 214                         hash = jhash2((u32 *)&flow.addrs.v6addrs, 8, hashrnd);
 215                 else
 216                         hash = 0;
 217
 218                 skb_set_hash(skb, hash, PKT_HASH_TYPE_L3);
 219         }
 220
 221         return hash;
 222 }
 223
 224 static inline int netvsc_get_tx_queue(struct net_device *ndev,
 225                                       struct sk_buff *skb, int old_idx)
 226 {
 227         const struct net_device_context *ndc = netdev_priv(ndev);
 228         struct sock *sk = skb->sk;
 229         int q_idx;
 230
 231         q_idx = ndc->tx_send_table[netvsc_get_hash(skb, sk) &
 232                                    (VRSS_SEND_TAB_SIZE - 1)];
 233
 234         /* If queue index changed record the new value */
 235         if (q_idx != old_idx &&
 236             sk && sk_fullsock(sk) && rcu_access_pointer(sk->sk_dst_cache))
 237                 sk_tx_queue_set(sk, q_idx);
 238
 239         return q_idx;
 240 }
 241
 242 /*
 243  * Select queue for transmit.
 244  *
 245  * If a valid queue has already been assigned, then use that.
 246  * Otherwise compute tx queue based on hash and the send table.
 247  *
 248  * This is basically similar to default (__netdev_pick_tx) with the added step
 249  * of using the host send_table when no other queue has been assigned.
 250  *
 251  * TODO support XPS - but get_xps_queue not exported
 252  */
 253 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
 254                         void *accel_priv, select_queue_fallback_t fallback)
 255 {
 256         unsigned int num_tx_queues = ndev->real_num_tx_queues;
 257         int q_idx = sk_tx_queue_get(skb->sk);
 258
 259         if (q_idx < 0 || skb->ooo_okay) {
 260                 /* If forwarding a packet, we use the recorded queue when
 261                  * available for better cache locality.
 262                  */
 263                 if (skb_rx_queue_recorded(skb))
 264                         q_idx = skb_get_rx_queue(skb);
 265                 else
 266                         q_idx = netvsc_get_tx_queue(ndev, skb, q_idx);
 267         }
 268
 269         while (unlikely(q_idx >= num_tx_queues))
 270                 q_idx -= num_tx_queues;
 271
 272         return q_idx;
 273 }
 274
 275 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
 276                         struct hv_page_buffer *pb)
 277 {
 278         int j = 0;
 279
 280         /* Deal with compund pages by ignoring unused part
 281          * of the page.
 282          */
 283         page += (offset >> PAGE_SHIFT);
 284         offset &= ~PAGE_MASK;
 285
 286         while (len > 0) {
 287                 unsigned long bytes;
 288
 289                 bytes = PAGE_SIZE - offset;
 290                 if (bytes > len)
 291                         bytes = len;
 292                 pb[j].pfn = page_to_pfn(page);
 293                 pb[j].offset = offset;
 294                 pb[j].len = bytes;
 295
 296                 offset += bytes;
 297                 len -= bytes;
 298
 299                 if (offset == PAGE_SIZE && len) {
 300                         page++;
 301                         offset = 0;
 302                         j++;
 303                 }
 304         }
 305
 306         return j + 1;
 307 }
 308
 309 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
 310                            struct hv_netvsc_packet *packet,
 311                            struct hv_page_buffer **page_buf)
 312 {
 313         struct hv_page_buffer *pb = *page_buf;
 314         u32 slots_used = 0;
 315         char *data = skb->data;
 316         int frags = skb_shinfo(skb)->nr_frags;
 317         int i;
 318
 319         /* The packet is laid out thus:
 320          * 1. hdr: RNDIS header and PPI
 321          * 2. skb linear data
 322          * 3. skb fragment data
 323          */
 324         if (hdr != NULL)
 325                 slots_used += fill_pg_buf(virt_to_page(hdr),
 326                                         offset_in_page(hdr),
 327                                         len, &pb[slots_used]);
 328
 329         packet->rmsg_size = len;
 330         packet->rmsg_pgcnt = slots_used;
 331
 332         slots_used += fill_pg_buf(virt_to_page(data),
 333                                 offset_in_page(data),
 334                                 skb_headlen(skb), &pb[slots_used]);
 335
 336         for (i = 0; i < frags; i++) {
 337                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 338
 339                 slots_used += fill_pg_buf(skb_frag_page(frag),
 340                                         frag->page_offset,
 341                                         skb_frag_size(frag), &pb[slots_used]);
 342         }
 343         return slots_used;
 344 }
 345
 346 static int count_skb_frag_slots(struct sk_buff *skb)
 347 {
 348         int i, frags = skb_shinfo(skb)->nr_frags;
 349         int pages = 0;
 350
 351         for (i = 0; i < frags; i++) {
 352                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 353                 unsigned long size = skb_frag_size(frag);
 354                 unsigned long offset = frag->page_offset;
 355
 356                 /* Skip unused frames from start of page */
 357                 offset &= ~PAGE_MASK;
 358                 pages += PFN_UP(offset + size);
 359         }
 360         return pages;
 361 }
 362
 363 static int netvsc_get_slots(struct sk_buff *skb)
 364 {
 365         char *data = skb->data;
 366         unsigned int offset = offset_in_page(data);
 367         unsigned int len = skb_headlen(skb);
 368         int slots;
 369         int frag_slots;
 370
 371         slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
 372         frag_slots = count_skb_frag_slots(skb);
 373         return slots + frag_slots;
 374 }
 375
 376 static u32 net_checksum_info(struct sk_buff *skb)
 377 {
 378         if (skb->protocol == htons(ETH_P_IP)) {
 379                 struct iphdr *ip = ip_hdr(skb);
 380
 381                 if (ip->protocol == IPPROTO_TCP)
 382                         return TRANSPORT_INFO_IPV4_TCP;
 383                 else if (ip->protocol == IPPROTO_UDP)
 384                         return TRANSPORT_INFO_IPV4_UDP;
 385         } else {
 386                 struct ipv6hdr *ip6 = ipv6_hdr(skb);
 387
 388                 if (ip6->nexthdr == IPPROTO_TCP)
 389                         return TRANSPORT_INFO_IPV6_TCP;
 390                 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
 391                         return TRANSPORT_INFO_IPV6_UDP;
 392         }
 393
 394         return TRANSPORT_INFO_NOT_IP;
 395 }
 396
 397 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
 398 {
 399         struct net_device_context *net_device_ctx = netdev_priv(net);
 400         struct hv_netvsc_packet *packet = NULL;
 401         int ret;
 402         unsigned int num_data_pgs;
 403         struct rndis_message *rndis_msg;
 404         struct rndis_packet *rndis_pkt;
 405         u32 rndis_msg_size;
 406         struct rndis_per_packet_info *ppi;
 407         u32 hash;
 408         struct hv_page_buffer page_buf[MAX_PAGE_BUFFER_COUNT];
 409         struct hv_page_buffer *pb = page_buf;
 410
 411         /* We will atmost need two pages to describe the rndis
 412          * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
 413          * of pages in a single packet. If skb is scattered around
 414          * more pages we try linearizing it.
 415          */
 416
 417         num_data_pgs = netvsc_get_slots(skb) + 2;
 418
 419         if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
 420                 ++net_device_ctx->eth_stats.tx_scattered;
 421
 422                 if (skb_linearize(skb))
 423                         goto no_memory;
 424
 425                 num_data_pgs = netvsc_get_slots(skb) + 2;
 426                 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
 427                         ++net_device_ctx->eth_stats.tx_too_big;
 428                         goto drop;
 429                 }
 430         }
 431
 432         /*
 433          * Place the rndis header in the skb head room and
 434          * the skb->cb will be used for hv_netvsc_packet
 435          * structure.
 436          */
 437         ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
 438         if (ret)
 439                 goto no_memory;
 440
 441         /* Use the skb control buffer for building up the packet */
 442         BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
 443                         FIELD_SIZEOF(struct sk_buff, cb));
 444         packet = (struct hv_netvsc_packet *)skb->cb;
 445
 446         packet->q_idx = skb_get_queue_mapping(skb);
 447
 448         packet->total_data_buflen = skb->len;
 449         packet->total_bytes = skb->len;
 450         packet->total_packets = 1;
 451
 452         rndis_msg = (struct rndis_message *)skb->head;
 453
 454         memset(rndis_msg, 0, RNDIS_AND_PPI_SIZE);
 455
 456         /* Add the rndis header */
 457         rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
 458         rndis_msg->msg_len = packet->total_data_buflen;
 459         rndis_pkt = &rndis_msg->msg.pkt;
 460         rndis_pkt->data_offset = sizeof(struct rndis_packet);
 461         rndis_pkt->data_len = packet->total_data_buflen;
 462         rndis_pkt->per_pkt_info_offset = sizeof(struct rndis_packet);
 463
 464         rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
 465
 466         hash = skb_get_hash_raw(skb);
 467         if (hash != 0 && net->real_num_tx_queues > 1) {
 468                 rndis_msg_size += NDIS_HASH_PPI_SIZE;
 469                 ppi = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
 470                                     NBL_HASH_VALUE);
 471                 *(u32 *)((void *)ppi + ppi->ppi_offset) = hash;
 472         }
 473
 474         if (skb_vlan_tag_present(skb)) {
 475                 struct ndis_pkt_8021q_info *vlan;
 476
 477                 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
 478                 ppi = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
 479                                         IEEE_8021Q_INFO);
 480                 vlan = (struct ndis_pkt_8021q_info *)((void *)ppi +
 481                                                 ppi->ppi_offset);
 482                 vlan->vlanid = skb->vlan_tci & VLAN_VID_MASK;
 483                 vlan->pri = (skb->vlan_tci & VLAN_PRIO_MASK) >>
 484                                 VLAN_PRIO_SHIFT;
 485         }
 486
 487         if (skb_is_gso(skb)) {
 488                 struct ndis_tcp_lso_info *lso_info;
 489
 490                 rndis_msg_size += NDIS_LSO_PPI_SIZE;
 491                 ppi = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
 492                                     TCP_LARGESEND_PKTINFO);
 493
 494                 lso_info = (struct ndis_tcp_lso_info *)((void *)ppi +
 495                                                         ppi->ppi_offset);
 496
 497                 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
 498                 if (skb->protocol == htons(ETH_P_IP)) {
 499                         lso_info->lso_v2_transmit.ip_version =
 500                                 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
 501                         ip_hdr(skb)->tot_len = 0;
 502                         ip_hdr(skb)->check = 0;
 503                         tcp_hdr(skb)->check =
 504                                 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
 505                                                    ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 506                 } else {
 507                         lso_info->lso_v2_transmit.ip_version =
 508                                 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
 509                         ipv6_hdr(skb)->payload_len = 0;
 510                         tcp_hdr(skb)->check =
 511                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 512                                                  &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 513                 }
 514                 lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb);
 515                 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
 516         } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
 517                 if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) {
 518                         struct ndis_tcp_ip_checksum_info *csum_info;
 519
 520                         rndis_msg_size += NDIS_CSUM_PPI_SIZE;
 521                         ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
 522                                             TCPIP_CHKSUM_PKTINFO);
 523
 524                         csum_info = (struct ndis_tcp_ip_checksum_info *)((void *)ppi +
 525                                                                          ppi->ppi_offset);
 526
 527                         csum_info->transmit.tcp_header_offset = skb_transport_offset(skb);
 528
 529                         if (skb->protocol == htons(ETH_P_IP)) {
 530                                 csum_info->transmit.is_ipv4 = 1;
 531
 532                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
 533                                         csum_info->transmit.tcp_checksum = 1;
 534                                 else
 535                                         csum_info->transmit.udp_checksum = 1;
 536                         } else {
 537                                 csum_info->transmit.is_ipv6 = 1;
 538
 539                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
 540                                         csum_info->transmit.tcp_checksum = 1;
 541                                 else
 542                                         csum_info->transmit.udp_checksum = 1;
 543                         }
 544                 } else {
 545                         /* Can't do offload of this type of checksum */
 546                         if (skb_checksum_help(skb))
 547                                 goto drop;
 548                 }
 549         }
 550
 551         /* Start filling in the page buffers with the rndis hdr */
 552         rndis_msg->msg_len += rndis_msg_size;
 553         packet->total_data_buflen = rndis_msg->msg_len;
 554         packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
 555                                                skb, packet, &pb);
 556
 557         /* timestamp packet in software */
 558         skb_tx_timestamp(skb);
 559         ret = netvsc_send(net_device_ctx->device_ctx, packet,
 560                           rndis_msg, &pb, skb);
 561         if (likely(ret == 0))
 562                 return NETDEV_TX_OK;
 563
 564         if (ret == -EAGAIN) {
 565                 ++net_device_ctx->eth_stats.tx_busy;
 566                 return NETDEV_TX_BUSY;
 567         }
 568
 569         if (ret == -ENOSPC)
 570                 ++net_device_ctx->eth_stats.tx_no_space;
 571
 572 drop:
 573         dev_kfree_skb_any(skb);
 574         net->stats.tx_dropped++;
 575
 576         return NETDEV_TX_OK;
 577
 578 no_memory:
 579         ++net_device_ctx->eth_stats.tx_no_memory;
 580         goto drop;
 581 }
 582 /*
 583  * netvsc_linkstatus_callback - Link up/down notification
 584  */
 585 void netvsc_linkstatus_callback(struct hv_device *device_obj,
 586                                 struct rndis_message *resp)
 587 {
 588         struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
 589         struct net_device *net;
 590         struct net_device_context *ndev_ctx;
 591         struct netvsc_reconfig *event;
 592         unsigned long flags;
 593
 594         net = hv_get_drvdata(device_obj);
 595
 596         if (!net)
 597                 return;
 598
 599         ndev_ctx = netdev_priv(net);
 600
 601         /* Update the physical link speed when changing to another vSwitch */
 602         if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
 603                 u32 speed;
 604
 605                 speed = *(u32 *)((void *)indicate + indicate->
 606                                  status_buf_offset) / 10000;
 607                 ndev_ctx->speed = speed;
 608                 return;
 609         }
 610
 611         /* Handle these link change statuses below */
 612         if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
 613             indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
 614             indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
 615                 return;
 616
 617         if (net->reg_state != NETREG_REGISTERED)
 618                 return;
 619
 620         event = kzalloc(sizeof(*event), GFP_ATOMIC);
 621         if (!event)
 622                 return;
 623         event->event = indicate->status;
 624
 625         spin_lock_irqsave(&ndev_ctx->lock, flags);
 626         list_add_tail(&event->list, &ndev_ctx->reconfig_events);
 627         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
 628
 629         schedule_delayed_work(&ndev_ctx->dwork, 0);
 630 }
 631
 632 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
 633                                              struct napi_struct *napi,
 634                                              const struct ndis_tcp_ip_checksum_info *csum_info,
 635                                              const struct ndis_pkt_8021q_info *vlan,
 636                                              void *data, u32 buflen)
 637 {
 638         struct sk_buff *skb;
 639
 640         skb = napi_alloc_skb(napi, buflen);
 641         if (!skb)
 642                 return skb;
 643
 644         /*
 645          * Copy to skb. This copy is needed here since the memory pointed by
 646          * hv_netvsc_packet cannot be deallocated
 647          */
 648         memcpy(skb_put(skb, buflen), data, buflen);
 649
 650         skb->protocol = eth_type_trans(skb, net);
 651
 652         /* skb is already created with CHECKSUM_NONE */
 653         skb_checksum_none_assert(skb);
 654
 655         /*
 656          * In Linux, the IP checksum is always checked.
 657          * Do L4 checksum offload if enabled and present.
 658          */
 659         if (csum_info && (net->features & NETIF_F_RXCSUM)) {
 660                 if (csum_info->receive.tcp_checksum_succeeded ||
 661                     csum_info->receive.udp_checksum_succeeded)
 662                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 663         }
 664
 665         if (vlan) {
 666                 u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT);
 667
 668                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 669                                        vlan_tci);
 670         }
 671
 672         return skb;
 673 }
 674
 675 /*
 676  * netvsc_recv_callback -  Callback when we receive a packet from the
 677  * "wire" on the specified device.
 678  */
 679 int netvsc_recv_callback(struct net_device *net,
 680                          struct vmbus_channel *channel,
 681                          void  *data, u32 len,
 682                          const struct ndis_tcp_ip_checksum_info *csum_info,
 683                          const struct ndis_pkt_8021q_info *vlan)
 684 {
 685         struct net_device_context *net_device_ctx = netdev_priv(net);
 686         struct netvsc_device *net_device;
 687         u16 q_idx = channel->offermsg.offer.sub_channel_index;
 688         struct netvsc_channel *nvchan;
 689         struct net_device *vf_netdev;
 690         struct sk_buff *skb;
 691         struct netvsc_stats *rx_stats;
 692
 693         if (net->reg_state != NETREG_REGISTERED)
 694                 return NVSP_STAT_FAIL;
 695
 696         /*
 697          * If necessary, inject this packet into the VF interface.
 698          * On Hyper-V, multicast and brodcast packets are only delivered
 699          * to the synthetic interface (after subjecting these to
 700          * policy filters on the host). Deliver these via the VF
 701          * interface in the guest.
 702          */
 703         rcu_read_lock();
 704         net_device = rcu_dereference(net_device_ctx->nvdev);
 705         if (unlikely(!net_device))
 706                 goto drop;
 707
 708         nvchan = &net_device->chan_table[q_idx];
 709         vf_netdev = rcu_dereference(net_device_ctx->vf_netdev);
 710         if (vf_netdev && (vf_netdev->flags & IFF_UP))
 711                 net = vf_netdev;
 712
 713         /* Allocate a skb - TODO direct I/O to pages? */
 714         skb = netvsc_alloc_recv_skb(net, &nvchan->napi,
 715                                     csum_info, vlan, data, len);
 716         if (unlikely(!skb)) {
 717 drop:
 718                 ++net->stats.rx_dropped;
 719                 rcu_read_unlock();
 720                 return NVSP_STAT_FAIL;
 721         }
 722
 723         if (net != vf_netdev)
 724                 skb_record_rx_queue(skb, q_idx);
 725
 726         /*
 727          * Even if injecting the packet, record the statistics
 728          * on the synthetic device because modifying the VF device
 729          * statistics will not work correctly.
 730          */
 731         rx_stats = &nvchan->rx_stats;
 732         u64_stats_update_begin(&rx_stats->syncp);
 733         rx_stats->packets++;
 734         rx_stats->bytes += len;
 735
 736         if (skb->pkt_type == PACKET_BROADCAST)
 737                 ++rx_stats->broadcast;
 738         else if (skb->pkt_type == PACKET_MULTICAST)
 739                 ++rx_stats->multicast;
 740         u64_stats_update_end(&rx_stats->syncp);
 741
 742         napi_gro_receive(&nvchan->napi, skb);
 743         rcu_read_unlock();
 744
 745         return 0;
 746 }
 747
 748 static void netvsc_get_drvinfo(struct net_device *net,
 749                                struct ethtool_drvinfo *info)
 750 {
 751         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
 752         strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
 753 }
 754
 755 static void netvsc_get_channels(struct net_device *net,
 756                                 struct ethtool_channels *channel)
 757 {
 758         struct net_device_context *net_device_ctx = netdev_priv(net);
 759         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
 760
 761         if (nvdev) {
 762                 channel->max_combined   = nvdev->max_chn;
 763                 channel->combined_count = nvdev->num_chn;
 764         }
 765 }
 766
 767 static int netvsc_set_queues(struct net_device *net, struct hv_device *dev,
 768                              u32 num_chn)
 769 {
 770         struct netvsc_device_info device_info;
 771         int ret;
 772
 773         memset(&device_info, 0, sizeof(device_info));
 774         device_info.num_chn = num_chn;
 775         device_info.ring_size = ring_size;
 776         device_info.max_num_vrss_chns = num_chn;
 777
 778         ret = rndis_filter_device_add(dev, &device_info);
 779         if (ret)
 780                 return ret;
 781
 782         ret = netif_set_real_num_tx_queues(net, num_chn);
 783         if (ret)
 784                 return ret;
 785
 786         ret = netif_set_real_num_rx_queues(net, num_chn);
 787
 788         return ret;
 789 }
 790
 791 static int netvsc_set_channels(struct net_device *net,
 792                                struct ethtool_channels *channels)
 793 {
 794         struct net_device_context *net_device_ctx = netdev_priv(net);
 795         struct hv_device *dev = net_device_ctx->device_ctx;
 796         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
 797         unsigned int count = channels->combined_count;
 798         bool was_running;
 799         int ret;
 800
 801         /* We do not support separate count for rx, tx, or other */
 802         if (count == 0 ||
 803             channels->rx_count || channels->tx_count || channels->other_count)
 804                 return -EINVAL;
 805
 806         if (count > net->num_tx_queues || count > net->num_rx_queues)
 807                 return -EINVAL;
 808
 809         if (!nvdev || nvdev->destroy)
 810                 return -ENODEV;
 811
 812         if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
 813                 return -EINVAL;
 814
 815         if (count > nvdev->max_chn)
 816                 return -EINVAL;
 817
 818         was_running = netif_running(net);
 819         if (was_running) {
 820                 ret = netvsc_close(net);
 821                 if (ret)
 822                         return ret;
 823         }
 824
 825         rndis_filter_device_remove(dev, nvdev);
 826
 827         ret = netvsc_set_queues(net, dev, count);
 828         if (ret == 0)
 829                 nvdev->num_chn = count;
 830         else
 831                 netvsc_set_queues(net, dev, nvdev->num_chn);
 832
 833         if (was_running)
 834                 ret = netvsc_open(net);
 835
 836         /* We may have missed link change notifications */
 837         schedule_delayed_work(&net_device_ctx->dwork, 0);
 838
 839         return ret;
 840 }
 841
 842 static bool
 843 netvsc_validate_ethtool_ss_cmd(const struct ethtool_link_ksettings *cmd)
 844 {
 845         struct ethtool_link_ksettings diff1 = *cmd;
 846         struct ethtool_link_ksettings diff2 = {};
 847
 848         diff1.base.speed = 0;
 849         diff1.base.duplex = 0;
 850         /* advertising and cmd are usually set */
 851         ethtool_link_ksettings_zero_link_mode(&diff1, advertising);
 852         diff1.base.cmd = 0;
 853         /* We set port to PORT_OTHER */
 854         diff2.base.port = PORT_OTHER;
 855
 856         return !memcmp(&diff1, &diff2, sizeof(diff1));
 857 }
 858
 859 static void netvsc_init_settings(struct net_device *dev)
 860 {
 861         struct net_device_context *ndc = netdev_priv(dev);
 862
 863         ndc->speed = SPEED_UNKNOWN;
 864         ndc->duplex = DUPLEX_FULL;
 865 }
 866
 867 static int netvsc_get_link_ksettings(struct net_device *dev,
 868                                      struct ethtool_link_ksettings *cmd)
 869 {
 870         struct net_device_context *ndc = netdev_priv(dev);
 871
 872         cmd->base.speed = ndc->speed;
 873         cmd->base.duplex = ndc->duplex;
 874         cmd->base.port = PORT_OTHER;
 875
 876         return 0;
 877 }
 878
 879 static int netvsc_set_link_ksettings(struct net_device *dev,
 880                                      const struct ethtool_link_ksettings *cmd)
 881 {
 882         struct net_device_context *ndc = netdev_priv(dev);
 883         u32 speed;
 884
 885         speed = cmd->base.speed;
 886         if (!ethtool_validate_speed(speed) ||
 887             !ethtool_validate_duplex(cmd->base.duplex) ||
 888             !netvsc_validate_ethtool_ss_cmd(cmd))
 889                 return -EINVAL;
 890
 891         ndc->speed = speed;
 892         ndc->duplex = cmd->base.duplex;
 893
 894         return 0;
 895 }
 896
 897 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
 898 {
 899         struct net_device_context *ndevctx = netdev_priv(ndev);
 900         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
 901         struct hv_device *hdev = ndevctx->device_ctx;
 902         struct netvsc_device_info device_info;
 903         bool was_running;
 904         int ret = 0;
 905
 906         if (!nvdev || nvdev->destroy)
 907                 return -ENODEV;
 908
 909         was_running = netif_running(ndev);
 910         if (was_running) {
 911                 ret = netvsc_close(ndev);
 912                 if (ret)
 913                         return ret;
 914         }
 915
 916         memset(&device_info, 0, sizeof(device_info));
 917         device_info.ring_size = ring_size;
 918         device_info.num_chn = nvdev->num_chn;
 919         device_info.max_num_vrss_chns = nvdev->num_chn;
 920
 921         rndis_filter_device_remove(hdev, nvdev);
 922
 923         /* 'nvdev' has been freed in rndis_filter_device_remove() ->
 924          * netvsc_device_remove () -> free_netvsc_device().
 925          * We mustn't access it before it's re-created in
 926          * rndis_filter_device_add() -> netvsc_device_add().
 927          */
 928
 929         ndev->mtu = mtu;
 930
 931         rndis_filter_device_add(hdev, &device_info);
 932
 933         if (was_running)
 934                 ret = netvsc_open(ndev);
 935
 936         /* We may have missed link change notifications */
 937         schedule_delayed_work(&ndevctx->dwork, 0);
 938
 939         return ret;
 940 }
 941
 942 static void netvsc_get_stats64(struct net_device *net,
 943                                struct rtnl_link_stats64 *t)
 944 {
 945         struct net_device_context *ndev_ctx = netdev_priv(net);
 946         struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
 947         int i;
 948
 949         if (!nvdev)
 950                 return;
 951
 952         for (i = 0; i < nvdev->num_chn; i++) {
 953                 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
 954                 const struct netvsc_stats *stats;
 955                 u64 packets, bytes, multicast;
 956                 unsigned int start;
 957
 958                 stats = &nvchan->tx_stats;
 959                 do {
 960                         start = u64_stats_fetch_begin_irq(&stats->syncp);
 961                         packets = stats->packets;
 962                         bytes = stats->bytes;
 963                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
 964
 965                 t->tx_bytes     += bytes;
 966                 t->tx_packets   += packets;
 967
 968                 stats = &nvchan->rx_stats;
 969                 do {
 970                         start = u64_stats_fetch_begin_irq(&stats->syncp);
 971                         packets = stats->packets;
 972                         bytes = stats->bytes;
 973                         multicast = stats->multicast + stats->broadcast;
 974                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
 975
 976                 t->rx_bytes     += bytes;
 977                 t->rx_packets   += packets;
 978                 t->multicast    += multicast;
 979         }
 980
 981         t->tx_dropped   = net->stats.tx_dropped;
 982         t->tx_errors    = net->stats.tx_errors;
 983
 984         t->rx_dropped   = net->stats.rx_dropped;
 985         t->rx_errors    = net->stats.rx_errors;
 986 }
 987
 988 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
 989 {
 990         struct sockaddr *addr = p;
 991         char save_adr[ETH_ALEN];
 992         unsigned char save_aatype;
 993         int err;
 994
 995         memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
 996         save_aatype = ndev->addr_assign_type;
 997
 998         err = eth_mac_addr(ndev, p);
 999         if (err != 0)
1000                 return err;
1001
1002         err = rndis_filter_set_device_mac(ndev, addr->sa_data);
1003         if (err != 0) {
1004                 /* roll back to saved MAC */
1005                 memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
1006                 ndev->addr_assign_type = save_aatype;
1007         }
1008
1009         return err;
1010 }
1011
1012 static const struct {
1013         char name[ETH_GSTRING_LEN];
1014         u16 offset;
1015 } netvsc_stats[] = {
1016         { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
1017         { "tx_no_memory",  offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
1018         { "tx_no_space",  offsetof(struct netvsc_ethtool_stats, tx_no_space) },
1019         { "tx_too_big",   offsetof(struct netvsc_ethtool_stats, tx_too_big) },
1020         { "tx_busy",      offsetof(struct netvsc_ethtool_stats, tx_busy) },
1021 };
1022
1023 #define NETVSC_GLOBAL_STATS_LEN ARRAY_SIZE(netvsc_stats)
1024
1025 /* 4 statistics per queue (rx/tx packets/bytes) */
1026 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 4)
1027
1028 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
1029 {
1030         struct net_device_context *ndc = netdev_priv(dev);
1031         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1032
1033         if (!nvdev)
1034                 return -ENODEV;
1035
1036         switch (string_set) {
1037         case ETH_SS_STATS:
1038                 return NETVSC_GLOBAL_STATS_LEN + NETVSC_QUEUE_STATS_LEN(nvdev);
1039         default:
1040                 return -EINVAL;
1041         }
1042 }
1043
1044 static void netvsc_get_ethtool_stats(struct net_device *dev,
1045                                      struct ethtool_stats *stats, u64 *data)
1046 {
1047         struct net_device_context *ndc = netdev_priv(dev);
1048         struct netvsc_device *nvdev = rcu_dereference(ndc->nvdev);
1049         const void *nds = &ndc->eth_stats;
1050         const struct netvsc_stats *qstats;
1051         unsigned int start;
1052         u64 packets, bytes;
1053         int i, j;
1054
1055         if (!nvdev)
1056                 return;
1057
1058         for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
1059                 data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1060
1061         for (j = 0; j < nvdev->num_chn; j++) {
1062                 qstats = &nvdev->chan_table[j].tx_stats;
1063
1064                 do {
1065                         start = u64_stats_fetch_begin_irq(&qstats->syncp);
1066                         packets = qstats->packets;
1067                         bytes = qstats->bytes;
1068                 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1069                 data[i++] = packets;
1070                 data[i++] = bytes;
1071
1072                 qstats = &nvdev->chan_table[j].rx_stats;
1073                 do {
1074                         start = u64_stats_fetch_begin_irq(&qstats->syncp);
1075                         packets = qstats->packets;
1076                         bytes = qstats->bytes;
1077                 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1078                 data[i++] = packets;
1079                 data[i++] = bytes;
1080         }
1081 }
1082
1083 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1084 {
1085         struct net_device_context *ndc = netdev_priv(dev);
1086         struct netvsc_device *nvdev = rcu_dereference(ndc->nvdev);
1087         u8 *p = data;
1088         int i;
1089
1090         if (!nvdev)
1091                 return;
1092
1093         switch (stringset) {
1094         case ETH_SS_STATS:
1095                 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++)
1096                         memcpy(p + i * ETH_GSTRING_LEN,
1097                                netvsc_stats[i].name, ETH_GSTRING_LEN);
1098
1099                 p += i * ETH_GSTRING_LEN;
1100                 for (i = 0; i < nvdev->num_chn; i++) {
1101                         sprintf(p, "tx_queue_%u_packets", i);
1102                         p += ETH_GSTRING_LEN;
1103                         sprintf(p, "tx_queue_%u_bytes", i);
1104                         p += ETH_GSTRING_LEN;
1105                         sprintf(p, "rx_queue_%u_packets", i);
1106                         p += ETH_GSTRING_LEN;
1107                         sprintf(p, "rx_queue_%u_bytes", i);
1108                         p += ETH_GSTRING_LEN;
1109                 }
1110
1111                 break;
1112         }
1113 }
1114
1115 static int
1116 netvsc_get_rss_hash_opts(struct netvsc_device *nvdev,
1117                          struct ethtool_rxnfc *info)
1118 {
1119         info->data = RXH_IP_SRC | RXH_IP_DST;
1120
1121         switch (info->flow_type) {
1122         case TCP_V4_FLOW:
1123         case TCP_V6_FLOW:
1124                 info->data |= RXH_L4_B_0_1 | RXH_L4_B_2_3;
1125                 /* fallthrough */
1126         case UDP_V4_FLOW:
1127         case UDP_V6_FLOW:
1128         case IPV4_FLOW:
1129         case IPV6_FLOW:
1130                 break;
1131         default:
1132                 info->data = 0;
1133                 break;
1134         }
1135
1136         return 0;
1137 }
1138
1139 static int
1140 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1141                  u32 *rules)
1142 {
1143         struct net_device_context *ndc = netdev_priv(dev);
1144         struct netvsc_device *nvdev = rcu_dereference(ndc->nvdev);
1145
1146         if (!nvdev)
1147                 return -ENODEV;
1148
1149         switch (info->cmd) {
1150         case ETHTOOL_GRXRINGS:
1151                 info->data = nvdev->num_chn;
1152                 return 0;
1153
1154         case ETHTOOL_GRXFH:
1155                 return netvsc_get_rss_hash_opts(nvdev, info);
1156         }
1157         return -EOPNOTSUPP;
1158 }
1159
1160 #ifdef CONFIG_NET_POLL_CONTROLLER
1161 static void netvsc_poll_controller(struct net_device *dev)
1162 {
1163         struct net_device_context *ndc = netdev_priv(dev);
1164         struct netvsc_device *ndev;
1165         int i;
1166
1167         rcu_read_lock();
1168         ndev = rcu_dereference(ndc->nvdev);
1169         if (ndev) {
1170                 for (i = 0; i < ndev->num_chn; i++) {
1171                         struct netvsc_channel *nvchan = &ndev->chan_table[i];
1172
1173                         napi_schedule(&nvchan->napi);
1174                 }
1175         }
1176         rcu_read_unlock();
1177 }
1178 #endif
1179
1180 static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1181 {
1182         return NETVSC_HASH_KEYLEN;
1183 }
1184
1185 static u32 netvsc_rss_indir_size(struct net_device *dev)
1186 {
1187         return ITAB_NUM;
1188 }
1189
1190 static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key,
1191                            u8 *hfunc)
1192 {
1193         struct net_device_context *ndc = netdev_priv(dev);
1194         struct netvsc_device *ndev = rcu_dereference(ndc->nvdev);
1195         struct rndis_device *rndis_dev;
1196         int i;
1197
1198         if (!ndev)
1199                 return -ENODEV;
1200
1201         if (hfunc)
1202                 *hfunc = ETH_RSS_HASH_TOP;      /* Toeplitz */
1203
1204         rndis_dev = ndev->extension;
1205         if (indir) {
1206                 for (i = 0; i < ITAB_NUM; i++)
1207                         indir[i] = rndis_dev->ind_table[i];
1208         }
1209
1210         if (key)
1211                 memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1212
1213         return 0;
1214 }
1215
1216 static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
1217                            const u8 *key, const u8 hfunc)
1218 {
1219         struct net_device_context *ndc = netdev_priv(dev);
1220         struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1221         struct rndis_device *rndis_dev;
1222         int i;
1223
1224         if (!ndev)
1225                 return -ENODEV;
1226
1227         if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
1228                 return -EOPNOTSUPP;
1229
1230         rndis_dev = ndev->extension;
1231         if (indir) {
1232                 for (i = 0; i < ITAB_NUM; i++)
1233                         if (indir[i] >= dev->num_rx_queues)
1234                                 return -EINVAL;
1235
1236                 for (i = 0; i < ITAB_NUM; i++)
1237                         rndis_dev->ind_table[i] = indir[i];
1238         }
1239
1240         if (!key) {
1241                 if (!indir)
1242                         return 0;
1243
1244                 key = rndis_dev->rss_key;
1245         }
1246
1247         return rndis_filter_set_rss_param(rndis_dev, key, ndev->num_chn);
1248 }
1249
1250 static const struct ethtool_ops ethtool_ops = {
1251         .get_drvinfo    = netvsc_get_drvinfo,
1252         .get_link       = ethtool_op_get_link,
1253         .get_ethtool_stats = netvsc_get_ethtool_stats,
1254         .get_sset_count = netvsc_get_sset_count,
1255         .get_strings    = netvsc_get_strings,
1256         .get_channels   = netvsc_get_channels,
1257         .set_channels   = netvsc_set_channels,
1258         .get_ts_info    = ethtool_op_get_ts_info,
1259         .get_rxnfc      = netvsc_get_rxnfc,
1260         .get_rxfh_key_size = netvsc_get_rxfh_key_size,
1261         .get_rxfh_indir_size = netvsc_rss_indir_size,
1262         .get_rxfh       = netvsc_get_rxfh,
1263         .set_rxfh       = netvsc_set_rxfh,
1264         .get_link_ksettings = netvsc_get_link_ksettings,
1265         .set_link_ksettings = netvsc_set_link_ksettings,
1266 };
1267
1268 static const struct net_device_ops device_ops = {
1269         .ndo_open =                     netvsc_open,
1270         .ndo_stop =                     netvsc_close,
1271         .ndo_start_xmit =               netvsc_start_xmit,
1272         .ndo_set_rx_mode =              netvsc_set_multicast_list,
1273         .ndo_change_mtu =               netvsc_change_mtu,
1274         .ndo_validate_addr =            eth_validate_addr,
1275         .ndo_set_mac_address =          netvsc_set_mac_addr,
1276         .ndo_select_queue =             netvsc_select_queue,
1277         .ndo_get_stats64 =              netvsc_get_stats64,
1278 #ifdef CONFIG_NET_POLL_CONTROLLER
1279         .ndo_poll_controller =          netvsc_poll_controller,
1280 #endif
1281 };
1282
1283 /*
1284  * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
1285  * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
1286  * present send GARP packet to network peers with netif_notify_peers().
1287  */
1288 static void netvsc_link_change(struct work_struct *w)
1289 {
1290         struct net_device_context *ndev_ctx =
1291                 container_of(w, struct net_device_context, dwork.work);
1292         struct hv_device *device_obj = ndev_ctx->device_ctx;
1293         struct net_device *net = hv_get_drvdata(device_obj);
1294         struct netvsc_device *net_device;
1295         struct rndis_device *rdev;
1296         struct netvsc_reconfig *event = NULL;
1297         bool notify = false, reschedule = false;
1298         unsigned long flags, next_reconfig, delay;
1299
1300         rtnl_lock();
1301         net_device = rtnl_dereference(ndev_ctx->nvdev);
1302         if (!net_device)
1303                 goto out_unlock;
1304
1305         rdev = net_device->extension;
1306
1307         next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
1308         if (time_is_after_jiffies(next_reconfig)) {
1309                 /* link_watch only sends one notification with current state
1310                  * per second, avoid doing reconfig more frequently. Handle
1311                  * wrap around.
1312                  */
1313                 delay = next_reconfig - jiffies;
1314                 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
1315                 schedule_delayed_work(&ndev_ctx->dwork, delay);
1316                 goto out_unlock;
1317         }
1318         ndev_ctx->last_reconfig = jiffies;
1319
1320         spin_lock_irqsave(&ndev_ctx->lock, flags);
1321         if (!list_empty(&ndev_ctx->reconfig_events)) {
1322                 event = list_first_entry(&ndev_ctx->reconfig_events,
1323                                          struct netvsc_reconfig, list);
1324                 list_del(&event->list);
1325                 reschedule = !list_empty(&ndev_ctx->reconfig_events);
1326         }
1327         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1328
1329         if (!event)
1330                 goto out_unlock;
1331
1332         switch (event->event) {
1333                 /* Only the following events are possible due to the check in
1334                  * netvsc_linkstatus_callback()
1335                  */
1336         case RNDIS_STATUS_MEDIA_CONNECT:
1337                 if (rdev->link_state) {
1338                         rdev->link_state = false;
1339                         netif_carrier_on(net);
1340                         netif_tx_wake_all_queues(net);
1341                 } else {
1342                         notify = true;
1343                 }
1344                 kfree(event);
1345                 break;
1346         case RNDIS_STATUS_MEDIA_DISCONNECT:
1347                 if (!rdev->link_state) {
1348                         rdev->link_state = true;
1349                         netif_carrier_off(net);
1350                         netif_tx_stop_all_queues(net);
1351                 }
1352                 kfree(event);
1353                 break;
1354         case RNDIS_STATUS_NETWORK_CHANGE:
1355                 /* Only makes sense if carrier is present */
1356                 if (!rdev->link_state) {
1357                         rdev->link_state = true;
1358                         netif_carrier_off(net);
1359                         netif_tx_stop_all_queues(net);
1360                         event->event = RNDIS_STATUS_MEDIA_CONNECT;
1361                         spin_lock_irqsave(&ndev_ctx->lock, flags);
1362                         list_add(&event->list, &ndev_ctx->reconfig_events);
1363                         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1364                         reschedule = true;
1365                 }
1366                 break;
1367         }
1368
1369         rtnl_unlock();
1370
1371         if (notify)
1372                 netdev_notify_peers(net);
1373
1374         /* link_watch only sends one notification with current state per
1375          * second, handle next reconfig event in 2 seconds.
1376          */
1377         if (reschedule)
1378                 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1379
1380         return;
1381
1382 out_unlock:
1383         rtnl_unlock();
1384 }
1385
1386 static struct net_device *get_netvsc_bymac(const u8 *mac)
1387 {
1388         struct net_device *dev;
1389
1390         ASSERT_RTNL();
1391
1392         for_each_netdev(&init_net, dev) {
1393                 if (dev->netdev_ops != &device_ops)
1394                         continue;       /* not a netvsc device */
1395
1396                 if (ether_addr_equal(mac, dev->perm_addr))
1397                         return dev;
1398         }
1399
1400         return NULL;
1401 }
1402
1403 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
1404 {
1405         struct net_device *dev;
1406
1407         ASSERT_RTNL();
1408
1409         for_each_netdev(&init_net, dev) {
1410                 struct net_device_context *net_device_ctx;
1411
1412                 if (dev->netdev_ops != &device_ops)
1413                         continue;       /* not a netvsc device */
1414
1415                 net_device_ctx = netdev_priv(dev);
1416                 if (net_device_ctx->nvdev == NULL)
1417                         continue;       /* device is removed */
1418
1419                 if (rtnl_dereference(net_device_ctx->vf_netdev) == vf_netdev)
1420                         return dev;     /* a match */
1421         }
1422
1423         return NULL;
1424 }
1425
1426 static int netvsc_register_vf(struct net_device *vf_netdev)
1427 {
1428         struct net_device *ndev;
1429         struct net_device_context *net_device_ctx;
1430         struct netvsc_device *netvsc_dev;
1431
1432         if (vf_netdev->addr_len != ETH_ALEN)
1433                 return NOTIFY_DONE;
1434
1435         /*
1436          * We will use the MAC address to locate the synthetic interface to
1437          * associate with the VF interface. If we don't find a matching
1438          * synthetic interface, move on.
1439          */
1440         ndev = get_netvsc_bymac(vf_netdev->perm_addr);
1441         if (!ndev)
1442                 return NOTIFY_DONE;
1443
1444         net_device_ctx = netdev_priv(ndev);
1445         netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
1446         if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
1447                 return NOTIFY_DONE;
1448
1449         netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
1450         /*
1451          * Take a reference on the module.
1452          */
1453         try_module_get(THIS_MODULE);
1454
1455         dev_hold(vf_netdev);
1456         rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
1457         return NOTIFY_OK;
1458 }
1459
1460 static int netvsc_vf_up(struct net_device *vf_netdev)
1461 {
1462         struct net_device *ndev;
1463         struct netvsc_device *netvsc_dev;
1464         struct net_device_context *net_device_ctx;
1465
1466         ndev = get_netvsc_byref(vf_netdev);
1467         if (!ndev)
1468                 return NOTIFY_DONE;
1469
1470         net_device_ctx = netdev_priv(ndev);
1471         netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
1472
1473         netdev_info(ndev, "VF up: %s\n", vf_netdev->name);
1474
1475         /*
1476          * Open the device before switching data path.
1477          */
1478         rndis_filter_open(netvsc_dev);
1479
1480         /*
1481          * notify the host to switch the data path.
1482          */
1483         netvsc_switch_datapath(ndev, true);
1484         netdev_info(ndev, "Data path switched to VF: %s\n", vf_netdev->name);
1485
1486         netif_carrier_off(ndev);
1487
1488         /* Now notify peers through VF device. */
1489         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, vf_netdev);
1490
1491         return NOTIFY_OK;
1492 }
1493
1494 static int netvsc_vf_down(struct net_device *vf_netdev)
1495 {
1496         struct net_device *ndev;
1497         struct netvsc_device *netvsc_dev;
1498         struct net_device_context *net_device_ctx;
1499
1500         ndev = get_netvsc_byref(vf_netdev);
1501         if (!ndev)
1502                 return NOTIFY_DONE;
1503
1504         net_device_ctx = netdev_priv(ndev);
1505         netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
1506
1507         netdev_info(ndev, "VF down: %s\n", vf_netdev->name);
1508         netvsc_switch_datapath(ndev, false);
1509         netdev_info(ndev, "Data path switched from VF: %s\n", vf_netdev->name);
1510         rndis_filter_close(netvsc_dev);
1511         netif_carrier_on(ndev);
1512
1513         /* Now notify peers through netvsc device. */
1514         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, ndev);
1515
1516         return NOTIFY_OK;
1517 }
1518
1519 static int netvsc_unregister_vf(struct net_device *vf_netdev)
1520 {
1521         struct net_device *ndev;
1522         struct net_device_context *net_device_ctx;
1523
1524         ndev = get_netvsc_byref(vf_netdev);
1525         if (!ndev)
1526                 return NOTIFY_DONE;
1527
1528         net_device_ctx = netdev_priv(ndev);
1529
1530         netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
1531
1532         RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
1533         dev_put(vf_netdev);
1534         module_put(THIS_MODULE);
1535         return NOTIFY_OK;
1536 }
1537
1538 static int netvsc_probe(struct hv_device *dev,
1539                         const struct hv_vmbus_device_id *dev_id)
1540 {
1541         struct net_device *net = NULL;
1542         struct net_device_context *net_device_ctx;
1543         struct netvsc_device_info device_info;
1544         struct netvsc_device *nvdev;
1545         int ret;
1546
1547         net = alloc_etherdev_mq(sizeof(struct net_device_context),
1548                                 VRSS_CHANNEL_MAX);
1549         if (!net)
1550                 return -ENOMEM;
1551
1552         netif_carrier_off(net);
1553
1554         netvsc_init_settings(net);
1555
1556         net_device_ctx = netdev_priv(net);
1557         net_device_ctx->device_ctx = dev;
1558         net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
1559         if (netif_msg_probe(net_device_ctx))
1560                 netdev_dbg(net, "netvsc msg_enable: %d\n",
1561                            net_device_ctx->msg_enable);
1562
1563         hv_set_drvdata(dev, net);
1564
1565         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
1566         INIT_WORK(&net_device_ctx->work, do_set_multicast);
1567
1568         spin_lock_init(&net_device_ctx->lock);
1569         INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
1570
1571         net->netdev_ops = &device_ops;
1572         net->ethtool_ops = &ethtool_ops;
1573         SET_NETDEV_DEV(net, &dev->device);
1574
1575         /* We always need headroom for rndis header */
1576         net->needed_headroom = RNDIS_AND_PPI_SIZE;
1577
1578         /* Notify the netvsc driver of the new device */
1579         memset(&device_info, 0, sizeof(device_info));
1580         device_info.ring_size = ring_size;
1581         device_info.num_chn = VRSS_CHANNEL_DEFAULT;
1582         ret = rndis_filter_device_add(dev, &device_info);
1583         if (ret != 0) {
1584                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
1585                 free_netdev(net);
1586                 hv_set_drvdata(dev, NULL);
1587                 return ret;
1588         }
1589         memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
1590
1591         /* hw_features computed in rndis_filter_device_add */
1592         net->features = net->hw_features |
1593                 NETIF_F_HIGHDMA | NETIF_F_SG |
1594                 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
1595         net->vlan_features = net->features;
1596
1597         /* RCU not necessary here, device not registered */
1598         nvdev = net_device_ctx->nvdev;
1599         netif_set_real_num_tx_queues(net, nvdev->num_chn);
1600         netif_set_real_num_rx_queues(net, nvdev->num_chn);
1601
1602         /* MTU range: 68 - 1500 or 65521 */
1603         net->min_mtu = NETVSC_MTU_MIN;
1604         if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
1605                 net->max_mtu = NETVSC_MTU - ETH_HLEN;
1606         else
1607                 net->max_mtu = ETH_DATA_LEN;
1608
1609         ret = register_netdev(net);
1610         if (ret != 0) {
1611                 pr_err("Unable to register netdev.\n");
1612                 rndis_filter_device_remove(dev, nvdev);
1613                 free_netdev(net);
1614         }
1615
1616         return ret;
1617 }
1618
1619 static int netvsc_remove(struct hv_device *dev)
1620 {
1621         struct net_device *net;
1622         struct net_device_context *ndev_ctx;
1623
1624         net = hv_get_drvdata(dev);
1625
1626         if (net == NULL) {
1627                 dev_err(&dev->device, "No net device to remove\n");
1628                 return 0;
1629         }
1630
1631         ndev_ctx = netdev_priv(net);
1632
1633         netif_device_detach(net);
1634
1635         cancel_delayed_work_sync(&ndev_ctx->dwork);
1636         cancel_work_sync(&ndev_ctx->work);
1637
1638         /*
1639          * Call to the vsc driver to let it know that the device is being
1640          * removed. Also blocks mtu and channel changes.
1641          */
1642         rtnl_lock();
1643         rndis_filter_device_remove(dev, ndev_ctx->nvdev);
1644         rtnl_unlock();
1645
1646         unregister_netdev(net);
1647
1648         hv_set_drvdata(dev, NULL);
1649
1650         free_netdev(net);
1651         return 0;
1652 }
1653
1654 static const struct hv_vmbus_device_id id_table[] = {
1655         /* Network guid */
1656         { HV_NIC_GUID, },
1657         { },
1658 };
1659
1660 MODULE_DEVICE_TABLE(vmbus, id_table);
1661
1662 /* The one and only one */
1663 static struct  hv_driver netvsc_drv = {
1664         .name = KBUILD_MODNAME,
1665         .id_table = id_table,
1666         .probe = netvsc_probe,
1667         .remove = netvsc_remove,
1668 };
1669
1670 /*
1671  * On Hyper-V, every VF interface is matched with a corresponding
1672  * synthetic interface. The synthetic interface is presented first
1673  * to the guest. When the corresponding VF instance is registered,
1674  * we will take care of switching the data path.
1675  */
1676 static int netvsc_netdev_event(struct notifier_block *this,
1677                                unsigned long event, void *ptr)
1678 {
1679         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
1680
1681         /* Skip our own events */
1682         if (event_dev->netdev_ops == &device_ops)
1683                 return NOTIFY_DONE;
1684
1685         /* Avoid non-Ethernet type devices */
1686         if (event_dev->type != ARPHRD_ETHER)
1687                 return NOTIFY_DONE;
1688
1689         /* Avoid Vlan dev with same MAC registering as VF */
1690         if (is_vlan_dev(event_dev))
1691                 return NOTIFY_DONE;
1692
1693         /* Avoid Bonding master dev with same MAC registering as VF */
1694         if ((event_dev->priv_flags & IFF_BONDING) &&
1695             (event_dev->flags & IFF_MASTER))
1696                 return NOTIFY_DONE;
1697
1698         switch (event) {
1699         case NETDEV_REGISTER:
1700                 return netvsc_register_vf(event_dev);
1701         case NETDEV_UNREGISTER:
1702                 return netvsc_unregister_vf(event_dev);
1703         case NETDEV_UP:
1704                 return netvsc_vf_up(event_dev);
1705         case NETDEV_DOWN:
1706                 return netvsc_vf_down(event_dev);
1707         default:
1708                 return NOTIFY_DONE;
1709         }
1710 }
1711
1712 static struct notifier_block netvsc_netdev_notifier = {
1713         .notifier_call = netvsc_netdev_event,
1714 };
1715
1716 static void __exit netvsc_drv_exit(void)
1717 {
1718         unregister_netdevice_notifier(&netvsc_netdev_notifier);
1719         vmbus_driver_unregister(&netvsc_drv);
1720 }
1721
1722 static int __init netvsc_drv_init(void)
1723 {
1724         int ret;
1725
1726         if (ring_size < RING_SIZE_MIN) {
1727                 ring_size = RING_SIZE_MIN;
1728                 pr_info("Increased ring_size to %d (min allowed)\n",
1729                         ring_size);
1730         }
1731         ret = vmbus_driver_register(&netvsc_drv);
1732
1733         if (ret)
1734                 return ret;
1735
1736         register_netdevice_notifier(&netvsc_netdev_notifier);
1737         return 0;
1738 }
1739
1740 MODULE_LICENSE("GPL");
1741 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
1742
1743 module_init(netvsc_drv_init);
1744 module_exit(netvsc_drv_exit);