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 <linux/rtnetlink.h>
  37 #include <linux/netpoll.h>
  38 #include <linux/reciprocal_div.h>
  39
  40 #include <net/arp.h>
  41 #include <net/route.h>
  42 #include <net/sock.h>
  43 #include <net/pkt_sched.h>
  44 #include <net/checksum.h>
  45 #include <net/ip6_checksum.h>
  46
  47 #include "hyperv_net.h"
  48
  49 #define RING_SIZE_MIN           64
  50
  51 #define LINKCHANGE_INT (2 * HZ)
  52 #define VF_TAKEOVER_INT (HZ / 10)
  53
  54 static unsigned int ring_size __ro_after_init = 128;
  55 module_param(ring_size, uint, S_IRUGO);
  56 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
  57 unsigned int netvsc_ring_bytes __ro_after_init;
  58 struct reciprocal_value netvsc_ring_reciprocal __ro_after_init;
  59
  60 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
  61                                 NETIF_MSG_LINK | NETIF_MSG_IFUP |
  62                                 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
  63                                 NETIF_MSG_TX_ERR;
  64
  65 static int debug = -1;
  66 module_param(debug, int, S_IRUGO);
  67 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  68
  69 static void netvsc_change_rx_flags(struct net_device *net, int change)
  70 {
  71         struct net_device_context *ndev_ctx = netdev_priv(net);
  72         struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
  73         int inc;
  74
  75         if (!vf_netdev)
  76                 return;
  77
  78         if (change & IFF_PROMISC) {
  79                 inc = (net->flags & IFF_PROMISC) ? 1 : -1;
  80                 dev_set_promiscuity(vf_netdev, inc);
  81         }
  82
  83         if (change & IFF_ALLMULTI) {
  84                 inc = (net->flags & IFF_ALLMULTI) ? 1 : -1;
  85                 dev_set_allmulti(vf_netdev, inc);
  86         }
  87 }
  88
  89 static void netvsc_set_rx_mode(struct net_device *net)
  90 {
  91         struct net_device_context *ndev_ctx = netdev_priv(net);
  92         struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
  93         struct netvsc_device *nvdev = rtnl_dereference(ndev_ctx->nvdev);
  94
  95         if (vf_netdev) {
  96                 dev_uc_sync(vf_netdev, net);
  97                 dev_mc_sync(vf_netdev, net);
  98         }
  99
 100         rndis_filter_update(nvdev);
 101 }
 102
 103 static int netvsc_open(struct net_device *net)
 104 {
 105         struct net_device_context *ndev_ctx = netdev_priv(net);
 106         struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
 107         struct netvsc_device *nvdev = rtnl_dereference(ndev_ctx->nvdev);
 108         struct rndis_device *rdev;
 109         int ret = 0;
 110
 111         netif_carrier_off(net);
 112
 113         /* Open up the device */
 114         ret = rndis_filter_open(nvdev);
 115         if (ret != 0) {
 116                 netdev_err(net, "unable to open device (ret %d).\n", ret);
 117                 return ret;
 118         }
 119
 120         rdev = nvdev->extension;
 121         if (!rdev->link_state) {
 122                 netif_carrier_on(net);
 123                 netif_tx_wake_all_queues(net);
 124         }
 125
 126         if (vf_netdev) {
 127                 /* Setting synthetic device up transparently sets
 128                  * slave as up. If open fails, then slave will be
 129                  * still be offline (and not used).
 130                  */
 131                 ret = dev_open(vf_netdev);
 132                 if (ret)
 133                         netdev_warn(net,
 134                                     "unable to open slave: %s: %d\n",
 135                                     vf_netdev->name, ret);
 136         }
 137         return 0;
 138 }
 139
 140 static int netvsc_close(struct net_device *net)
 141 {
 142         struct net_device_context *net_device_ctx = netdev_priv(net);
 143         struct net_device *vf_netdev
 144                 = rtnl_dereference(net_device_ctx->vf_netdev);
 145         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
 146         int ret = 0;
 147         u32 aread, i, msec = 10, retry = 0, retry_max = 20;
 148         struct vmbus_channel *chn;
 149
 150         netif_tx_disable(net);
 151
 152         /* No need to close rndis filter if it is removed already */
 153         if (!nvdev)
 154                 goto out;
 155
 156         ret = rndis_filter_close(nvdev);
 157         if (ret != 0) {
 158                 netdev_err(net, "unable to close device (ret %d).\n", ret);
 159                 return ret;
 160         }
 161
 162         /* Ensure pending bytes in ring are read */
 163         while (true) {
 164                 aread = 0;
 165                 for (i = 0; i < nvdev->num_chn; i++) {
 166                         chn = nvdev->chan_table[i].channel;
 167                         if (!chn)
 168                                 continue;
 169
 170                         aread = hv_get_bytes_to_read(&chn->inbound);
 171                         if (aread)
 172                                 break;
 173
 174                         aread = hv_get_bytes_to_read(&chn->outbound);
 175                         if (aread)
 176                                 break;
 177                 }
 178
 179                 retry++;
 180                 if (retry > retry_max || aread == 0)
 181                         break;
 182
 183                 msleep(msec);
 184
 185                 if (msec < 1000)
 186                         msec *= 2;
 187         }
 188
 189         if (aread) {
 190                 netdev_err(net, "Ring buffer not empty after closing rndis\n");
 191                 ret = -ETIMEDOUT;
 192         }
 193
 194 out:
 195         if (vf_netdev)
 196                 dev_close(vf_netdev);
 197
 198         return ret;
 199 }
 200
 201 static inline void *init_ppi_data(struct rndis_message *msg,
 202                                   u32 ppi_size, u32 pkt_type)
 203 {
 204         struct rndis_packet *rndis_pkt = &msg->msg.pkt;
 205         struct rndis_per_packet_info *ppi;
 206
 207         rndis_pkt->data_offset += ppi_size;
 208         ppi = (void *)rndis_pkt + rndis_pkt->per_pkt_info_offset
 209                 + rndis_pkt->per_pkt_info_len;
 210
 211         ppi->size = ppi_size;
 212         ppi->type = pkt_type;
 213         ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
 214
 215         rndis_pkt->per_pkt_info_len += ppi_size;
 216
 217         return ppi + 1;
 218 }
 219
 220 /* Azure hosts don't support non-TCP port numbers in hashing for fragmented
 221  * packets. We can use ethtool to change UDP hash level when necessary.
 222  */
 223 static inline u32 netvsc_get_hash(
 224         struct sk_buff *skb,
 225         const struct net_device_context *ndc)
 226 {
 227         struct flow_keys flow;
 228         u32 hash, pkt_proto = 0;
 229         static u32 hashrnd __read_mostly;
 230
 231         net_get_random_once(&hashrnd, sizeof(hashrnd));
 232
 233         if (!skb_flow_dissect_flow_keys(skb, &flow, 0))
 234                 return 0;
 235
 236         switch (flow.basic.ip_proto) {
 237         case IPPROTO_TCP:
 238                 if (flow.basic.n_proto == htons(ETH_P_IP))
 239                         pkt_proto = HV_TCP4_L4HASH;
 240                 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
 241                         pkt_proto = HV_TCP6_L4HASH;
 242
 243                 break;
 244
 245         case IPPROTO_UDP:
 246                 if (flow.basic.n_proto == htons(ETH_P_IP))
 247                         pkt_proto = HV_UDP4_L4HASH;
 248                 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
 249                         pkt_proto = HV_UDP6_L4HASH;
 250
 251                 break;
 252         }
 253
 254         if (pkt_proto & ndc->l4_hash) {
 255                 return skb_get_hash(skb);
 256         } else {
 257                 if (flow.basic.n_proto == htons(ETH_P_IP))
 258                         hash = jhash2((u32 *)&flow.addrs.v4addrs, 2, hashrnd);
 259                 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
 260                         hash = jhash2((u32 *)&flow.addrs.v6addrs, 8, hashrnd);
 261                 else
 262                         hash = 0;
 263
 264                 skb_set_hash(skb, hash, PKT_HASH_TYPE_L3);
 265         }
 266
 267         return hash;
 268 }
 269
 270 static inline int netvsc_get_tx_queue(struct net_device *ndev,
 271                                       struct sk_buff *skb, int old_idx)
 272 {
 273         const struct net_device_context *ndc = netdev_priv(ndev);
 274         struct sock *sk = skb->sk;
 275         int q_idx;
 276
 277         q_idx = ndc->tx_table[netvsc_get_hash(skb, ndc) &
 278                               (VRSS_SEND_TAB_SIZE - 1)];
 279
 280         /* If queue index changed record the new value */
 281         if (q_idx != old_idx &&
 282             sk && sk_fullsock(sk) && rcu_access_pointer(sk->sk_dst_cache))
 283                 sk_tx_queue_set(sk, q_idx);
 284
 285         return q_idx;
 286 }
 287
 288 /*
 289  * Select queue for transmit.
 290  *
 291  * If a valid queue has already been assigned, then use that.
 292  * Otherwise compute tx queue based on hash and the send table.
 293  *
 294  * This is basically similar to default (__netdev_pick_tx) with the added step
 295  * of using the host send_table when no other queue has been assigned.
 296  *
 297  * TODO support XPS - but get_xps_queue not exported
 298  */
 299 static u16 netvsc_pick_tx(struct net_device *ndev, struct sk_buff *skb)
 300 {
 301         int q_idx = sk_tx_queue_get(skb->sk);
 302
 303         if (q_idx < 0 || skb->ooo_okay || q_idx >= ndev->real_num_tx_queues) {
 304                 /* If forwarding a packet, we use the recorded queue when
 305                  * available for better cache locality.
 306                  */
 307                 if (skb_rx_queue_recorded(skb))
 308                         q_idx = skb_get_rx_queue(skb);
 309                 else
 310                         q_idx = netvsc_get_tx_queue(ndev, skb, q_idx);
 311         }
 312
 313         return q_idx;
 314 }
 315
 316 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
 317                                void *accel_priv,
 318                                select_queue_fallback_t fallback)
 319 {
 320         struct net_device_context *ndc = netdev_priv(ndev);
 321         struct net_device *vf_netdev;
 322         u16 txq;
 323
 324         rcu_read_lock();
 325         vf_netdev = rcu_dereference(ndc->vf_netdev);
 326         if (vf_netdev) {
 327                 const struct net_device_ops *vf_ops = vf_netdev->netdev_ops;
 328
 329                 if (vf_ops->ndo_select_queue)
 330                         txq = vf_ops->ndo_select_queue(vf_netdev, skb,
 331                                                        accel_priv, fallback);
 332                 else
 333                         txq = fallback(vf_netdev, skb);
 334
 335                 /* Record the queue selected by VF so that it can be
 336                  * used for common case where VF has more queues than
 337                  * the synthetic device.
 338                  */
 339                 qdisc_skb_cb(skb)->slave_dev_queue_mapping = txq;
 340         } else {
 341                 txq = netvsc_pick_tx(ndev, skb);
 342         }
 343         rcu_read_unlock();
 344
 345         while (unlikely(txq >= ndev->real_num_tx_queues))
 346                 txq -= ndev->real_num_tx_queues;
 347
 348         return txq;
 349 }
 350
 351 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
 352                        struct hv_page_buffer *pb)
 353 {
 354         int j = 0;
 355
 356         /* Deal with compund pages by ignoring unused part
 357          * of the page.
 358          */
 359         page += (offset >> PAGE_SHIFT);
 360         offset &= ~PAGE_MASK;
 361
 362         while (len > 0) {
 363                 unsigned long bytes;
 364
 365                 bytes = PAGE_SIZE - offset;
 366                 if (bytes > len)
 367                         bytes = len;
 368                 pb[j].pfn = page_to_pfn(page);
 369                 pb[j].offset = offset;
 370                 pb[j].len = bytes;
 371
 372                 offset += bytes;
 373                 len -= bytes;
 374
 375                 if (offset == PAGE_SIZE && len) {
 376                         page++;
 377                         offset = 0;
 378                         j++;
 379                 }
 380         }
 381
 382         return j + 1;
 383 }
 384
 385 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
 386                            struct hv_netvsc_packet *packet,
 387                            struct hv_page_buffer *pb)
 388 {
 389         u32 slots_used = 0;
 390         char *data = skb->data;
 391         int frags = skb_shinfo(skb)->nr_frags;
 392         int i;
 393
 394         /* The packet is laid out thus:
 395          * 1. hdr: RNDIS header and PPI
 396          * 2. skb linear data
 397          * 3. skb fragment data
 398          */
 399         slots_used += fill_pg_buf(virt_to_page(hdr),
 400                                   offset_in_page(hdr),
 401                                   len, &pb[slots_used]);
 402
 403         packet->rmsg_size = len;
 404         packet->rmsg_pgcnt = slots_used;
 405
 406         slots_used += fill_pg_buf(virt_to_page(data),
 407                                 offset_in_page(data),
 408                                 skb_headlen(skb), &pb[slots_used]);
 409
 410         for (i = 0; i < frags; i++) {
 411                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 412
 413                 slots_used += fill_pg_buf(skb_frag_page(frag),
 414                                         frag->page_offset,
 415                                         skb_frag_size(frag), &pb[slots_used]);
 416         }
 417         return slots_used;
 418 }
 419
 420 static int count_skb_frag_slots(struct sk_buff *skb)
 421 {
 422         int i, frags = skb_shinfo(skb)->nr_frags;
 423         int pages = 0;
 424
 425         for (i = 0; i < frags; i++) {
 426                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 427                 unsigned long size = skb_frag_size(frag);
 428                 unsigned long offset = frag->page_offset;
 429
 430                 /* Skip unused frames from start of page */
 431                 offset &= ~PAGE_MASK;
 432                 pages += PFN_UP(offset + size);
 433         }
 434         return pages;
 435 }
 436
 437 static int netvsc_get_slots(struct sk_buff *skb)
 438 {
 439         char *data = skb->data;
 440         unsigned int offset = offset_in_page(data);
 441         unsigned int len = skb_headlen(skb);
 442         int slots;
 443         int frag_slots;
 444
 445         slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
 446         frag_slots = count_skb_frag_slots(skb);
 447         return slots + frag_slots;
 448 }
 449
 450 static u32 net_checksum_info(struct sk_buff *skb)
 451 {
 452         if (skb->protocol == htons(ETH_P_IP)) {
 453                 struct iphdr *ip = ip_hdr(skb);
 454
 455                 if (ip->protocol == IPPROTO_TCP)
 456                         return TRANSPORT_INFO_IPV4_TCP;
 457                 else if (ip->protocol == IPPROTO_UDP)
 458                         return TRANSPORT_INFO_IPV4_UDP;
 459         } else {
 460                 struct ipv6hdr *ip6 = ipv6_hdr(skb);
 461
 462                 if (ip6->nexthdr == IPPROTO_TCP)
 463                         return TRANSPORT_INFO_IPV6_TCP;
 464                 else if (ip6->nexthdr == IPPROTO_UDP)
 465                         return TRANSPORT_INFO_IPV6_UDP;
 466         }
 467
 468         return TRANSPORT_INFO_NOT_IP;
 469 }
 470
 471 /* Send skb on the slave VF device. */
 472 static int netvsc_vf_xmit(struct net_device *net, struct net_device *vf_netdev,
 473                           struct sk_buff *skb)
 474 {
 475         struct net_device_context *ndev_ctx = netdev_priv(net);
 476         unsigned int len = skb->len;
 477         int rc;
 478
 479         skb->dev = vf_netdev;
 480         skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
 481
 482         rc = dev_queue_xmit(skb);
 483         if (likely(rc == NET_XMIT_SUCCESS || rc == NET_XMIT_CN)) {
 484                 struct netvsc_vf_pcpu_stats *pcpu_stats
 485                         = this_cpu_ptr(ndev_ctx->vf_stats);
 486
 487                 u64_stats_update_begin(&pcpu_stats->syncp);
 488                 pcpu_stats->tx_packets++;
 489                 pcpu_stats->tx_bytes += len;
 490                 u64_stats_update_end(&pcpu_stats->syncp);
 491         } else {
 492                 this_cpu_inc(ndev_ctx->vf_stats->tx_dropped);
 493         }
 494
 495         return rc;
 496 }
 497
 498 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
 499 {
 500         struct net_device_context *net_device_ctx = netdev_priv(net);
 501         struct hv_netvsc_packet *packet = NULL;
 502         int ret;
 503         unsigned int num_data_pgs;
 504         struct rndis_message *rndis_msg;
 505         struct net_device *vf_netdev;
 506         u32 rndis_msg_size;
 507         u32 hash;
 508         struct hv_page_buffer pb[MAX_PAGE_BUFFER_COUNT];
 509
 510         /* if VF is present and up then redirect packets
 511          * already called with rcu_read_lock_bh
 512          */
 513         vf_netdev = rcu_dereference_bh(net_device_ctx->vf_netdev);
 514         if (vf_netdev && netif_running(vf_netdev) &&
 515             !netpoll_tx_running(net))
 516                 return netvsc_vf_xmit(net, vf_netdev, skb);
 517
 518         /* We will atmost need two pages to describe the rndis
 519          * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
 520          * of pages in a single packet. If skb is scattered around
 521          * more pages we try linearizing it.
 522          */
 523
 524         num_data_pgs = netvsc_get_slots(skb) + 2;
 525
 526         if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
 527                 ++net_device_ctx->eth_stats.tx_scattered;
 528
 529                 if (skb_linearize(skb))
 530                         goto no_memory;
 531
 532                 num_data_pgs = netvsc_get_slots(skb) + 2;
 533                 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
 534                         ++net_device_ctx->eth_stats.tx_too_big;
 535                         goto drop;
 536                 }
 537         }
 538
 539         /*
 540          * Place the rndis header in the skb head room and
 541          * the skb->cb will be used for hv_netvsc_packet
 542          * structure.
 543          */
 544         ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
 545         if (ret)
 546                 goto no_memory;
 547
 548         /* Use the skb control buffer for building up the packet */
 549         BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
 550                         FIELD_SIZEOF(struct sk_buff, cb));
 551         packet = (struct hv_netvsc_packet *)skb->cb;
 552
 553         packet->q_idx = skb_get_queue_mapping(skb);
 554
 555         packet->total_data_buflen = skb->len;
 556         packet->total_bytes = skb->len;
 557         packet->total_packets = 1;
 558
 559         rndis_msg = (struct rndis_message *)skb->head;
 560
 561         /* Add the rndis header */
 562         rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
 563         rndis_msg->msg_len = packet->total_data_buflen;
 564
 565         rndis_msg->msg.pkt = (struct rndis_packet) {
 566                 .data_offset = sizeof(struct rndis_packet),
 567                 .data_len = packet->total_data_buflen,
 568                 .per_pkt_info_offset = sizeof(struct rndis_packet),
 569         };
 570
 571         rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
 572
 573         hash = skb_get_hash_raw(skb);
 574         if (hash != 0 && net->real_num_tx_queues > 1) {
 575                 u32 *hash_info;
 576
 577                 rndis_msg_size += NDIS_HASH_PPI_SIZE;
 578                 hash_info = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
 579                                           NBL_HASH_VALUE);
 580                 *hash_info = hash;
 581         }
 582
 583         if (skb_vlan_tag_present(skb)) {
 584                 struct ndis_pkt_8021q_info *vlan;
 585
 586                 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
 587                 vlan = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
 588                                      IEEE_8021Q_INFO);
 589
 590                 vlan->value = 0;
 591                 vlan->vlanid = skb->vlan_tci & VLAN_VID_MASK;
 592                 vlan->pri = (skb->vlan_tci & VLAN_PRIO_MASK) >>
 593                                 VLAN_PRIO_SHIFT;
 594         }
 595
 596         if (skb_is_gso(skb)) {
 597                 struct ndis_tcp_lso_info *lso_info;
 598
 599                 rndis_msg_size += NDIS_LSO_PPI_SIZE;
 600                 lso_info = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
 601                                          TCP_LARGESEND_PKTINFO);
 602
 603                 lso_info->value = 0;
 604                 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
 605                 if (skb->protocol == htons(ETH_P_IP)) {
 606                         lso_info->lso_v2_transmit.ip_version =
 607                                 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
 608                         ip_hdr(skb)->tot_len = 0;
 609                         ip_hdr(skb)->check = 0;
 610                         tcp_hdr(skb)->check =
 611                                 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
 612                                                    ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 613                 } else {
 614                         lso_info->lso_v2_transmit.ip_version =
 615                                 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
 616                         ipv6_hdr(skb)->payload_len = 0;
 617                         tcp_hdr(skb)->check =
 618                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 619                                                  &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 620                 }
 621                 lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb);
 622                 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
 623         } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
 624                 if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) {
 625                         struct ndis_tcp_ip_checksum_info *csum_info;
 626
 627                         rndis_msg_size += NDIS_CSUM_PPI_SIZE;
 628                         csum_info = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
 629                                                   TCPIP_CHKSUM_PKTINFO);
 630
 631                         csum_info->value = 0;
 632                         csum_info->transmit.tcp_header_offset = skb_transport_offset(skb);
 633
 634                         if (skb->protocol == htons(ETH_P_IP)) {
 635                                 csum_info->transmit.is_ipv4 = 1;
 636
 637                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
 638                                         csum_info->transmit.tcp_checksum = 1;
 639                                 else
 640                                         csum_info->transmit.udp_checksum = 1;
 641                         } else {
 642                                 csum_info->transmit.is_ipv6 = 1;
 643
 644                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
 645                                         csum_info->transmit.tcp_checksum = 1;
 646                                 else
 647                                         csum_info->transmit.udp_checksum = 1;
 648                         }
 649                 } else {
 650                         /* Can't do offload of this type of checksum */
 651                         if (skb_checksum_help(skb))
 652                                 goto drop;
 653                 }
 654         }
 655
 656         /* Start filling in the page buffers with the rndis hdr */
 657         rndis_msg->msg_len += rndis_msg_size;
 658         packet->total_data_buflen = rndis_msg->msg_len;
 659         packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
 660                                                skb, packet, pb);
 661
 662         /* timestamp packet in software */
 663         skb_tx_timestamp(skb);
 664
 665         ret = netvsc_send(net, packet, rndis_msg, pb, skb);
 666         if (likely(ret == 0))
 667                 return NETDEV_TX_OK;
 668
 669         if (ret == -EAGAIN) {
 670                 ++net_device_ctx->eth_stats.tx_busy;
 671                 return NETDEV_TX_BUSY;
 672         }
 673
 674         if (ret == -ENOSPC)
 675                 ++net_device_ctx->eth_stats.tx_no_space;
 676
 677 drop:
 678         dev_kfree_skb_any(skb);
 679         net->stats.tx_dropped++;
 680
 681         return NETDEV_TX_OK;
 682
 683 no_memory:
 684         ++net_device_ctx->eth_stats.tx_no_memory;
 685         goto drop;
 686 }
 687
 688 /*
 689  * netvsc_linkstatus_callback - Link up/down notification
 690  */
 691 void netvsc_linkstatus_callback(struct net_device *net,
 692                                 struct rndis_message *resp)
 693 {
 694         struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
 695         struct net_device_context *ndev_ctx = netdev_priv(net);
 696         struct netvsc_reconfig *event;
 697         unsigned long flags;
 698
 699         /* Update the physical link speed when changing to another vSwitch */
 700         if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
 701                 u32 speed;
 702
 703                 speed = *(u32 *)((void *)indicate
 704                                  + indicate->status_buf_offset) / 10000;
 705                 ndev_ctx->speed = speed;
 706                 return;
 707         }
 708
 709         /* Handle these link change statuses below */
 710         if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
 711             indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
 712             indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
 713                 return;
 714
 715         if (net->reg_state != NETREG_REGISTERED)
 716                 return;
 717
 718         event = kzalloc(sizeof(*event), GFP_ATOMIC);
 719         if (!event)
 720                 return;
 721         event->event = indicate->status;
 722
 723         spin_lock_irqsave(&ndev_ctx->lock, flags);
 724         list_add_tail(&event->list, &ndev_ctx->reconfig_events);
 725         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
 726
 727         schedule_delayed_work(&ndev_ctx->dwork, 0);
 728 }
 729
 730 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
 731                                              struct napi_struct *napi,
 732                                              const struct ndis_tcp_ip_checksum_info *csum_info,
 733                                              const struct ndis_pkt_8021q_info *vlan,
 734                                              void *data, u32 buflen)
 735 {
 736         struct sk_buff *skb;
 737
 738         skb = napi_alloc_skb(napi, buflen);
 739         if (!skb)
 740                 return skb;
 741
 742         /*
 743          * Copy to skb. This copy is needed here since the memory pointed by
 744          * hv_netvsc_packet cannot be deallocated
 745          */
 746         skb_put_data(skb, data, buflen);
 747
 748         skb->protocol = eth_type_trans(skb, net);
 749
 750         /* skb is already created with CHECKSUM_NONE */
 751         skb_checksum_none_assert(skb);
 752
 753         /*
 754          * In Linux, the IP checksum is always checked.
 755          * Do L4 checksum offload if enabled and present.
 756          */
 757         if (csum_info && (net->features & NETIF_F_RXCSUM)) {
 758                 if (csum_info->receive.tcp_checksum_succeeded ||
 759                     csum_info->receive.udp_checksum_succeeded)
 760                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 761         }
 762
 763         if (vlan) {
 764                 u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT);
 765
 766                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 767                                        vlan_tci);
 768         }
 769
 770         return skb;
 771 }
 772
 773 /*
 774  * netvsc_recv_callback -  Callback when we receive a packet from the
 775  * "wire" on the specified device.
 776  */
 777 int netvsc_recv_callback(struct net_device *net,
 778                          struct netvsc_device *net_device,
 779                          struct vmbus_channel *channel,
 780                          void  *data, u32 len,
 781                          const struct ndis_tcp_ip_checksum_info *csum_info,
 782                          const struct ndis_pkt_8021q_info *vlan)
 783 {
 784         struct net_device_context *net_device_ctx = netdev_priv(net);
 785         u16 q_idx = channel->offermsg.offer.sub_channel_index;
 786         struct netvsc_channel *nvchan = &net_device->chan_table[q_idx];
 787         struct sk_buff *skb;
 788         struct netvsc_stats *rx_stats;
 789
 790         if (net->reg_state != NETREG_REGISTERED)
 791                 return NVSP_STAT_FAIL;
 792
 793         /* Allocate a skb - TODO direct I/O to pages? */
 794         skb = netvsc_alloc_recv_skb(net, &nvchan->napi,
 795                                     csum_info, vlan, data, len);
 796         if (unlikely(!skb)) {
 797                 ++net_device_ctx->eth_stats.rx_no_memory;
 798                 rcu_read_unlock();
 799                 return NVSP_STAT_FAIL;
 800         }
 801
 802         skb_record_rx_queue(skb, q_idx);
 803
 804         /*
 805          * Even if injecting the packet, record the statistics
 806          * on the synthetic device because modifying the VF device
 807          * statistics will not work correctly.
 808          */
 809         rx_stats = &nvchan->rx_stats;
 810         u64_stats_update_begin(&rx_stats->syncp);
 811         rx_stats->packets++;
 812         rx_stats->bytes += len;
 813
 814         if (skb->pkt_type == PACKET_BROADCAST)
 815                 ++rx_stats->broadcast;
 816         else if (skb->pkt_type == PACKET_MULTICAST)
 817                 ++rx_stats->multicast;
 818         u64_stats_update_end(&rx_stats->syncp);
 819
 820         napi_gro_receive(&nvchan->napi, skb);
 821         return 0;
 822 }
 823
 824 static void netvsc_get_drvinfo(struct net_device *net,
 825                                struct ethtool_drvinfo *info)
 826 {
 827         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
 828         strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
 829 }
 830
 831 static void netvsc_get_channels(struct net_device *net,
 832                                 struct ethtool_channels *channel)
 833 {
 834         struct net_device_context *net_device_ctx = netdev_priv(net);
 835         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
 836
 837         if (nvdev) {
 838                 channel->max_combined   = nvdev->max_chn;
 839                 channel->combined_count = nvdev->num_chn;
 840         }
 841 }
 842
 843 static int netvsc_set_channels(struct net_device *net,
 844                                struct ethtool_channels *channels)
 845 {
 846         struct net_device_context *net_device_ctx = netdev_priv(net);
 847         struct hv_device *dev = net_device_ctx->device_ctx;
 848         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
 849         unsigned int orig, count = channels->combined_count;
 850         struct netvsc_device_info device_info;
 851         bool was_opened;
 852         int ret = 0;
 853
 854         /* We do not support separate count for rx, tx, or other */
 855         if (count == 0 ||
 856             channels->rx_count || channels->tx_count || channels->other_count)
 857                 return -EINVAL;
 858
 859         if (!nvdev || nvdev->destroy)
 860                 return -ENODEV;
 861
 862         if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
 863                 return -EINVAL;
 864
 865         if (count > nvdev->max_chn)
 866                 return -EINVAL;
 867
 868         orig = nvdev->num_chn;
 869         was_opened = rndis_filter_opened(nvdev);
 870         if (was_opened)
 871                 rndis_filter_close(nvdev);
 872
 873         memset(&device_info, 0, sizeof(device_info));
 874         device_info.num_chn = count;
 875         device_info.send_sections = nvdev->send_section_cnt;
 876         device_info.send_section_size = nvdev->send_section_size;
 877         device_info.recv_sections = nvdev->recv_section_cnt;
 878         device_info.recv_section_size = nvdev->recv_section_size;
 879
 880         rndis_filter_device_remove(dev, nvdev);
 881
 882         nvdev = rndis_filter_device_add(dev, &device_info);
 883         if (IS_ERR(nvdev)) {
 884                 ret = PTR_ERR(nvdev);
 885                 device_info.num_chn = orig;
 886                 nvdev = rndis_filter_device_add(dev, &device_info);
 887
 888                 if (IS_ERR(nvdev)) {
 889                         netdev_err(net, "restoring channel setting failed: %ld\n",
 890                                    PTR_ERR(nvdev));
 891                         return ret;
 892                 }
 893         }
 894
 895         if (was_opened)
 896                 rndis_filter_open(nvdev);
 897
 898         /* We may have missed link change notifications */
 899         net_device_ctx->last_reconfig = 0;
 900         schedule_delayed_work(&net_device_ctx->dwork, 0);
 901
 902         return ret;
 903 }
 904
 905 static bool
 906 netvsc_validate_ethtool_ss_cmd(const struct ethtool_link_ksettings *cmd)
 907 {
 908         struct ethtool_link_ksettings diff1 = *cmd;
 909         struct ethtool_link_ksettings diff2 = {};
 910
 911         diff1.base.speed = 0;
 912         diff1.base.duplex = 0;
 913         /* advertising and cmd are usually set */
 914         ethtool_link_ksettings_zero_link_mode(&diff1, advertising);
 915         diff1.base.cmd = 0;
 916         /* We set port to PORT_OTHER */
 917         diff2.base.port = PORT_OTHER;
 918
 919         return !memcmp(&diff1, &diff2, sizeof(diff1));
 920 }
 921
 922 static void netvsc_init_settings(struct net_device *dev)
 923 {
 924         struct net_device_context *ndc = netdev_priv(dev);
 925
 926         ndc->l4_hash = HV_DEFAULT_L4HASH;
 927
 928         ndc->speed = SPEED_UNKNOWN;
 929         ndc->duplex = DUPLEX_FULL;
 930 }
 931
 932 static int netvsc_get_link_ksettings(struct net_device *dev,
 933                                      struct ethtool_link_ksettings *cmd)
 934 {
 935         struct net_device_context *ndc = netdev_priv(dev);
 936
 937         cmd->base.speed = ndc->speed;
 938         cmd->base.duplex = ndc->duplex;
 939         cmd->base.port = PORT_OTHER;
 940
 941         return 0;
 942 }
 943
 944 static int netvsc_set_link_ksettings(struct net_device *dev,
 945                                      const struct ethtool_link_ksettings *cmd)
 946 {
 947         struct net_device_context *ndc = netdev_priv(dev);
 948         u32 speed;
 949
 950         speed = cmd->base.speed;
 951         if (!ethtool_validate_speed(speed) ||
 952             !ethtool_validate_duplex(cmd->base.duplex) ||
 953             !netvsc_validate_ethtool_ss_cmd(cmd))
 954                 return -EINVAL;
 955
 956         ndc->speed = speed;
 957         ndc->duplex = cmd->base.duplex;
 958
 959         return 0;
 960 }
 961
 962 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
 963 {
 964         struct net_device_context *ndevctx = netdev_priv(ndev);
 965         struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
 966         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
 967         struct hv_device *hdev = ndevctx->device_ctx;
 968         int orig_mtu = ndev->mtu;
 969         struct netvsc_device_info device_info;
 970         bool was_opened;
 971         int ret = 0;
 972
 973         if (!nvdev || nvdev->destroy)
 974                 return -ENODEV;
 975
 976         /* Change MTU of underlying VF netdev first. */
 977         if (vf_netdev) {
 978                 ret = dev_set_mtu(vf_netdev, mtu);
 979                 if (ret)
 980                         return ret;
 981         }
 982
 983         netif_device_detach(ndev);
 984         was_opened = rndis_filter_opened(nvdev);
 985         if (was_opened)
 986                 rndis_filter_close(nvdev);
 987
 988         memset(&device_info, 0, sizeof(device_info));
 989         device_info.num_chn = nvdev->num_chn;
 990         device_info.send_sections = nvdev->send_section_cnt;
 991         device_info.send_section_size = nvdev->send_section_size;
 992         device_info.recv_sections = nvdev->recv_section_cnt;
 993         device_info.recv_section_size = nvdev->recv_section_size;
 994
 995         rndis_filter_device_remove(hdev, nvdev);
 996
 997         ndev->mtu = mtu;
 998
 999         nvdev = rndis_filter_device_add(hdev, &device_info);
1000         if (IS_ERR(nvdev)) {
1001                 ret = PTR_ERR(nvdev);
1002
1003                 /* Attempt rollback to original MTU */
1004                 ndev->mtu = orig_mtu;
1005                 nvdev = rndis_filter_device_add(hdev, &device_info);
1006
1007                 if (vf_netdev)
1008                         dev_set_mtu(vf_netdev, orig_mtu);
1009
1010                 if (IS_ERR(nvdev)) {
1011                         netdev_err(ndev, "restoring mtu failed: %ld\n",
1012                                    PTR_ERR(nvdev));
1013                         return ret;
1014                 }
1015         }
1016
1017         if (was_opened)
1018                 rndis_filter_open(nvdev);
1019
1020         netif_device_attach(ndev);
1021
1022         /* We may have missed link change notifications */
1023         schedule_delayed_work(&ndevctx->dwork, 0);
1024
1025         return ret;
1026 }
1027
1028 static void netvsc_get_vf_stats(struct net_device *net,
1029                                 struct netvsc_vf_pcpu_stats *tot)
1030 {
1031         struct net_device_context *ndev_ctx = netdev_priv(net);
1032         int i;
1033
1034         memset(tot, 0, sizeof(*tot));
1035
1036         for_each_possible_cpu(i) {
1037                 const struct netvsc_vf_pcpu_stats *stats
1038                         = per_cpu_ptr(ndev_ctx->vf_stats, i);
1039                 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1040                 unsigned int start;
1041
1042                 do {
1043                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1044                         rx_packets = stats->rx_packets;
1045                         tx_packets = stats->tx_packets;
1046                         rx_bytes = stats->rx_bytes;
1047                         tx_bytes = stats->tx_bytes;
1048                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1049
1050                 tot->rx_packets += rx_packets;
1051                 tot->tx_packets += tx_packets;
1052                 tot->rx_bytes   += rx_bytes;
1053                 tot->tx_bytes   += tx_bytes;
1054                 tot->tx_dropped += stats->tx_dropped;
1055         }
1056 }
1057
1058 static void netvsc_get_stats64(struct net_device *net,
1059                                struct rtnl_link_stats64 *t)
1060 {
1061         struct net_device_context *ndev_ctx = netdev_priv(net);
1062         struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
1063         struct netvsc_vf_pcpu_stats vf_tot;
1064         int i;
1065
1066         if (!nvdev)
1067                 return;
1068
1069         netdev_stats_to_stats64(t, &net->stats);
1070
1071         netvsc_get_vf_stats(net, &vf_tot);
1072         t->rx_packets += vf_tot.rx_packets;
1073         t->tx_packets += vf_tot.tx_packets;
1074         t->rx_bytes   += vf_tot.rx_bytes;
1075         t->tx_bytes   += vf_tot.tx_bytes;
1076         t->tx_dropped += vf_tot.tx_dropped;
1077
1078         for (i = 0; i < nvdev->num_chn; i++) {
1079                 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1080                 const struct netvsc_stats *stats;
1081                 u64 packets, bytes, multicast;
1082                 unsigned int start;
1083
1084                 stats = &nvchan->tx_stats;
1085                 do {
1086                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1087                         packets = stats->packets;
1088                         bytes = stats->bytes;
1089                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1090
1091                 t->tx_bytes     += bytes;
1092                 t->tx_packets   += packets;
1093
1094                 stats = &nvchan->rx_stats;
1095                 do {
1096                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1097                         packets = stats->packets;
1098                         bytes = stats->bytes;
1099                         multicast = stats->multicast + stats->broadcast;
1100                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1101
1102                 t->rx_bytes     += bytes;
1103                 t->rx_packets   += packets;
1104                 t->multicast    += multicast;
1105         }
1106 }
1107
1108 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
1109 {
1110         struct net_device_context *ndc = netdev_priv(ndev);
1111         struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1112         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1113         struct sockaddr *addr = p;
1114         int err;
1115
1116         err = eth_prepare_mac_addr_change(ndev, p);
1117         if (err)
1118                 return err;
1119
1120         if (!nvdev)
1121                 return -ENODEV;
1122
1123         if (vf_netdev) {
1124                 err = dev_set_mac_address(vf_netdev, addr);
1125                 if (err)
1126                         return err;
1127         }
1128
1129         err = rndis_filter_set_device_mac(nvdev, addr->sa_data);
1130         if (!err) {
1131                 eth_commit_mac_addr_change(ndev, p);
1132         } else if (vf_netdev) {
1133                 /* rollback change on VF */
1134                 memcpy(addr->sa_data, ndev->dev_addr, ETH_ALEN);
1135                 dev_set_mac_address(vf_netdev, addr);
1136         }
1137
1138         return err;
1139 }
1140
1141 static const struct {
1142         char name[ETH_GSTRING_LEN];
1143         u16 offset;
1144 } netvsc_stats[] = {
1145         { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
1146         { "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
1147         { "tx_no_space",  offsetof(struct netvsc_ethtool_stats, tx_no_space) },
1148         { "tx_too_big",   offsetof(struct netvsc_ethtool_stats, tx_too_big) },
1149         { "tx_busy",      offsetof(struct netvsc_ethtool_stats, tx_busy) },
1150         { "tx_send_full", offsetof(struct netvsc_ethtool_stats, tx_send_full) },
1151         { "rx_comp_busy", offsetof(struct netvsc_ethtool_stats, rx_comp_busy) },
1152         { "rx_no_memory", offsetof(struct netvsc_ethtool_stats, rx_no_memory) },
1153         { "stop_queue", offsetof(struct netvsc_ethtool_stats, stop_queue) },
1154         { "wake_queue", offsetof(struct netvsc_ethtool_stats, wake_queue) },
1155 }, vf_stats[] = {
1156         { "vf_rx_packets", offsetof(struct netvsc_vf_pcpu_stats, rx_packets) },
1157         { "vf_rx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, rx_bytes) },
1158         { "vf_tx_packets", offsetof(struct netvsc_vf_pcpu_stats, tx_packets) },
1159         { "vf_tx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, tx_bytes) },
1160         { "vf_tx_dropped", offsetof(struct netvsc_vf_pcpu_stats, tx_dropped) },
1161 };
1162
1163 #define NETVSC_GLOBAL_STATS_LEN ARRAY_SIZE(netvsc_stats)
1164 #define NETVSC_VF_STATS_LEN     ARRAY_SIZE(vf_stats)
1165
1166 /* 4 statistics per queue (rx/tx packets/bytes) */
1167 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 4)
1168
1169 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
1170 {
1171         struct net_device_context *ndc = netdev_priv(dev);
1172         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1173
1174         if (!nvdev)
1175                 return -ENODEV;
1176
1177         switch (string_set) {
1178         case ETH_SS_STATS:
1179                 return NETVSC_GLOBAL_STATS_LEN
1180                         + NETVSC_VF_STATS_LEN
1181                         + NETVSC_QUEUE_STATS_LEN(nvdev);
1182         default:
1183                 return -EINVAL;
1184         }
1185 }
1186
1187 static void netvsc_get_ethtool_stats(struct net_device *dev,
1188                                      struct ethtool_stats *stats, u64 *data)
1189 {
1190         struct net_device_context *ndc = netdev_priv(dev);
1191         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1192         const void *nds = &ndc->eth_stats;
1193         const struct netvsc_stats *qstats;
1194         struct netvsc_vf_pcpu_stats sum;
1195         unsigned int start;
1196         u64 packets, bytes;
1197         int i, j;
1198
1199         if (!nvdev)
1200                 return;
1201
1202         for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
1203                 data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1204
1205         netvsc_get_vf_stats(dev, &sum);
1206         for (j = 0; j < NETVSC_VF_STATS_LEN; j++)
1207                 data[i++] = *(u64 *)((void *)&sum + vf_stats[j].offset);
1208
1209         for (j = 0; j < nvdev->num_chn; j++) {
1210                 qstats = &nvdev->chan_table[j].tx_stats;
1211
1212                 do {
1213                         start = u64_stats_fetch_begin_irq(&qstats->syncp);
1214                         packets = qstats->packets;
1215                         bytes = qstats->bytes;
1216                 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1217                 data[i++] = packets;
1218                 data[i++] = bytes;
1219
1220                 qstats = &nvdev->chan_table[j].rx_stats;
1221                 do {
1222                         start = u64_stats_fetch_begin_irq(&qstats->syncp);
1223                         packets = qstats->packets;
1224                         bytes = qstats->bytes;
1225                 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1226                 data[i++] = packets;
1227                 data[i++] = bytes;
1228         }
1229 }
1230
1231 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1232 {
1233         struct net_device_context *ndc = netdev_priv(dev);
1234         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1235         u8 *p = data;
1236         int i;
1237
1238         if (!nvdev)
1239                 return;
1240
1241         switch (stringset) {
1242         case ETH_SS_STATS:
1243                 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++) {
1244                         memcpy(p, netvsc_stats[i].name, ETH_GSTRING_LEN);
1245                         p += ETH_GSTRING_LEN;
1246                 }
1247
1248                 for (i = 0; i < ARRAY_SIZE(vf_stats); i++) {
1249                         memcpy(p, vf_stats[i].name, ETH_GSTRING_LEN);
1250                         p += ETH_GSTRING_LEN;
1251                 }
1252
1253                 for (i = 0; i < nvdev->num_chn; i++) {
1254                         sprintf(p, "tx_queue_%u_packets", i);
1255                         p += ETH_GSTRING_LEN;
1256                         sprintf(p, "tx_queue_%u_bytes", i);
1257                         p += ETH_GSTRING_LEN;
1258                         sprintf(p, "rx_queue_%u_packets", i);
1259                         p += ETH_GSTRING_LEN;
1260                         sprintf(p, "rx_queue_%u_bytes", i);
1261                         p += ETH_GSTRING_LEN;
1262                 }
1263
1264                 break;
1265         }
1266 }
1267
1268 static int
1269 netvsc_get_rss_hash_opts(struct net_device_context *ndc,
1270                          struct ethtool_rxnfc *info)
1271 {
1272         const u32 l4_flag = RXH_L4_B_0_1 | RXH_L4_B_2_3;
1273
1274         info->data = RXH_IP_SRC | RXH_IP_DST;
1275
1276         switch (info->flow_type) {
1277         case TCP_V4_FLOW:
1278                 if (ndc->l4_hash & HV_TCP4_L4HASH)
1279                         info->data |= l4_flag;
1280
1281                 break;
1282
1283         case TCP_V6_FLOW:
1284                 if (ndc->l4_hash & HV_TCP6_L4HASH)
1285                         info->data |= l4_flag;
1286
1287                 break;
1288
1289         case UDP_V4_FLOW:
1290                 if (ndc->l4_hash & HV_UDP4_L4HASH)
1291                         info->data |= l4_flag;
1292
1293                 break;
1294
1295         case UDP_V6_FLOW:
1296                 if (ndc->l4_hash & HV_UDP6_L4HASH)
1297                         info->data |= l4_flag;
1298
1299                 break;
1300
1301         case IPV4_FLOW:
1302         case IPV6_FLOW:
1303                 break;
1304         default:
1305                 info->data = 0;
1306                 break;
1307         }
1308
1309         return 0;
1310 }
1311
1312 static int
1313 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1314                  u32 *rules)
1315 {
1316         struct net_device_context *ndc = netdev_priv(dev);
1317         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1318
1319         if (!nvdev)
1320                 return -ENODEV;
1321
1322         switch (info->cmd) {
1323         case ETHTOOL_GRXRINGS:
1324                 info->data = nvdev->num_chn;
1325                 return 0;
1326
1327         case ETHTOOL_GRXFH:
1328                 return netvsc_get_rss_hash_opts(ndc, info);
1329         }
1330         return -EOPNOTSUPP;
1331 }
1332
1333 static int netvsc_set_rss_hash_opts(struct net_device_context *ndc,
1334                                     struct ethtool_rxnfc *info)
1335 {
1336         if (info->data == (RXH_IP_SRC | RXH_IP_DST |
1337                            RXH_L4_B_0_1 | RXH_L4_B_2_3)) {
1338                 switch (info->flow_type) {
1339                 case TCP_V4_FLOW:
1340                         ndc->l4_hash |= HV_TCP4_L4HASH;
1341                         break;
1342
1343                 case TCP_V6_FLOW:
1344                         ndc->l4_hash |= HV_TCP6_L4HASH;
1345                         break;
1346
1347                 case UDP_V4_FLOW:
1348                         ndc->l4_hash |= HV_UDP4_L4HASH;
1349                         break;
1350
1351                 case UDP_V6_FLOW:
1352                         ndc->l4_hash |= HV_UDP6_L4HASH;
1353                         break;
1354
1355                 default:
1356                         return -EOPNOTSUPP;
1357                 }
1358
1359                 return 0;
1360         }
1361
1362         if (info->data == (RXH_IP_SRC | RXH_IP_DST)) {
1363                 switch (info->flow_type) {
1364                 case TCP_V4_FLOW:
1365                         ndc->l4_hash &= ~HV_TCP4_L4HASH;
1366                         break;
1367
1368                 case TCP_V6_FLOW:
1369                         ndc->l4_hash &= ~HV_TCP6_L4HASH;
1370                         break;
1371
1372                 case UDP_V4_FLOW:
1373                         ndc->l4_hash &= ~HV_UDP4_L4HASH;
1374                         break;
1375
1376                 case UDP_V6_FLOW:
1377                         ndc->l4_hash &= ~HV_UDP6_L4HASH;
1378                         break;
1379
1380                 default:
1381                         return -EOPNOTSUPP;
1382                 }
1383
1384                 return 0;
1385         }
1386
1387         return -EOPNOTSUPP;
1388 }
1389
1390 static int
1391 netvsc_set_rxnfc(struct net_device *ndev, struct ethtool_rxnfc *info)
1392 {
1393         struct net_device_context *ndc = netdev_priv(ndev);
1394
1395         if (info->cmd == ETHTOOL_SRXFH)
1396                 return netvsc_set_rss_hash_opts(ndc, info);
1397
1398         return -EOPNOTSUPP;
1399 }
1400
1401 #ifdef CONFIG_NET_POLL_CONTROLLER
1402 static void netvsc_poll_controller(struct net_device *dev)
1403 {
1404         struct net_device_context *ndc = netdev_priv(dev);
1405         struct netvsc_device *ndev;
1406         int i;
1407
1408         rcu_read_lock();
1409         ndev = rcu_dereference(ndc->nvdev);
1410         if (ndev) {
1411                 for (i = 0; i < ndev->num_chn; i++) {
1412                         struct netvsc_channel *nvchan = &ndev->chan_table[i];
1413
1414                         napi_schedule(&nvchan->napi);
1415                 }
1416         }
1417         rcu_read_unlock();
1418 }
1419 #endif
1420
1421 static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1422 {
1423         return NETVSC_HASH_KEYLEN;
1424 }
1425
1426 static u32 netvsc_rss_indir_size(struct net_device *dev)
1427 {
1428         return ITAB_NUM;
1429 }
1430
1431 static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key,
1432                            u8 *hfunc)
1433 {
1434         struct net_device_context *ndc = netdev_priv(dev);
1435         struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1436         struct rndis_device *rndis_dev;
1437         int i;
1438
1439         if (!ndev)
1440                 return -ENODEV;
1441
1442         if (hfunc)
1443                 *hfunc = ETH_RSS_HASH_TOP;      /* Toeplitz */
1444
1445         rndis_dev = ndev->extension;
1446         if (indir) {
1447                 for (i = 0; i < ITAB_NUM; i++)
1448                         indir[i] = rndis_dev->rx_table[i];
1449         }
1450
1451         if (key)
1452                 memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1453
1454         return 0;
1455 }
1456
1457 static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
1458                            const u8 *key, const u8 hfunc)
1459 {
1460         struct net_device_context *ndc = netdev_priv(dev);
1461         struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1462         struct rndis_device *rndis_dev;
1463         int i;
1464
1465         if (!ndev)
1466                 return -ENODEV;
1467
1468         if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
1469                 return -EOPNOTSUPP;
1470
1471         rndis_dev = ndev->extension;
1472         if (indir) {
1473                 for (i = 0; i < ITAB_NUM; i++)
1474                         if (indir[i] >= ndev->num_chn)
1475                                 return -EINVAL;
1476
1477                 for (i = 0; i < ITAB_NUM; i++)
1478                         rndis_dev->rx_table[i] = indir[i];
1479         }
1480
1481         if (!key) {
1482                 if (!indir)
1483                         return 0;
1484
1485                 key = rndis_dev->rss_key;
1486         }
1487
1488         return rndis_filter_set_rss_param(rndis_dev, key);
1489 }
1490
1491 /* Hyper-V RNDIS protocol does not have ring in the HW sense.
1492  * It does have pre-allocated receive area which is divided into sections.
1493  */
1494 static void __netvsc_get_ringparam(struct netvsc_device *nvdev,
1495                                    struct ethtool_ringparam *ring)
1496 {
1497         u32 max_buf_size;
1498
1499         ring->rx_pending = nvdev->recv_section_cnt;
1500         ring->tx_pending = nvdev->send_section_cnt;
1501
1502         if (nvdev->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
1503                 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
1504         else
1505                 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
1506
1507         ring->rx_max_pending = max_buf_size / nvdev->recv_section_size;
1508         ring->tx_max_pending = NETVSC_SEND_BUFFER_SIZE
1509                 / nvdev->send_section_size;
1510 }
1511
1512 static void netvsc_get_ringparam(struct net_device *ndev,
1513                                  struct ethtool_ringparam *ring)
1514 {
1515         struct net_device_context *ndevctx = netdev_priv(ndev);
1516         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1517
1518         if (!nvdev)
1519                 return;
1520
1521         __netvsc_get_ringparam(nvdev, ring);
1522 }
1523
1524 static int netvsc_set_ringparam(struct net_device *ndev,
1525                                 struct ethtool_ringparam *ring)
1526 {
1527         struct net_device_context *ndevctx = netdev_priv(ndev);
1528         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1529         struct hv_device *hdev = ndevctx->device_ctx;
1530         struct netvsc_device_info device_info;
1531         struct ethtool_ringparam orig;
1532         u32 new_tx, new_rx;
1533         bool was_opened;
1534         int ret = 0;
1535
1536         if (!nvdev || nvdev->destroy)
1537                 return -ENODEV;
1538
1539         memset(&orig, 0, sizeof(orig));
1540         __netvsc_get_ringparam(nvdev, &orig);
1541
1542         new_tx = clamp_t(u32, ring->tx_pending,
1543                          NETVSC_MIN_TX_SECTIONS, orig.tx_max_pending);
1544         new_rx = clamp_t(u32, ring->rx_pending,
1545                          NETVSC_MIN_RX_SECTIONS, orig.rx_max_pending);
1546
1547         if (new_tx == orig.tx_pending &&
1548             new_rx == orig.rx_pending)
1549                 return 0;        /* no change */
1550
1551         memset(&device_info, 0, sizeof(device_info));
1552         device_info.num_chn = nvdev->num_chn;
1553         device_info.send_sections = new_tx;
1554         device_info.send_section_size = nvdev->send_section_size;
1555         device_info.recv_sections = new_rx;
1556         device_info.recv_section_size = nvdev->recv_section_size;
1557
1558         netif_device_detach(ndev);
1559         was_opened = rndis_filter_opened(nvdev);
1560         if (was_opened)
1561                 rndis_filter_close(nvdev);
1562
1563         rndis_filter_device_remove(hdev, nvdev);
1564
1565         nvdev = rndis_filter_device_add(hdev, &device_info);
1566         if (IS_ERR(nvdev)) {
1567                 ret = PTR_ERR(nvdev);
1568
1569                 device_info.send_sections = orig.tx_pending;
1570                 device_info.recv_sections = orig.rx_pending;
1571                 nvdev = rndis_filter_device_add(hdev, &device_info);
1572                 if (IS_ERR(nvdev)) {
1573                         netdev_err(ndev, "restoring ringparam failed: %ld\n",
1574                                    PTR_ERR(nvdev));
1575                         return ret;
1576                 }
1577         }
1578
1579         if (was_opened)
1580                 rndis_filter_open(nvdev);
1581         netif_device_attach(ndev);
1582
1583         /* We may have missed link change notifications */
1584         ndevctx->last_reconfig = 0;
1585         schedule_delayed_work(&ndevctx->dwork, 0);
1586
1587         return ret;
1588 }
1589
1590 static const struct ethtool_ops ethtool_ops = {
1591         .get_drvinfo    = netvsc_get_drvinfo,
1592         .get_link       = ethtool_op_get_link,
1593         .get_ethtool_stats = netvsc_get_ethtool_stats,
1594         .get_sset_count = netvsc_get_sset_count,
1595         .get_strings    = netvsc_get_strings,
1596         .get_channels   = netvsc_get_channels,
1597         .set_channels   = netvsc_set_channels,
1598         .get_ts_info    = ethtool_op_get_ts_info,
1599         .get_rxnfc      = netvsc_get_rxnfc,
1600         .set_rxnfc      = netvsc_set_rxnfc,
1601         .get_rxfh_key_size = netvsc_get_rxfh_key_size,
1602         .get_rxfh_indir_size = netvsc_rss_indir_size,
1603         .get_rxfh       = netvsc_get_rxfh,
1604         .set_rxfh       = netvsc_set_rxfh,
1605         .get_link_ksettings = netvsc_get_link_ksettings,
1606         .set_link_ksettings = netvsc_set_link_ksettings,
1607         .get_ringparam  = netvsc_get_ringparam,
1608         .set_ringparam  = netvsc_set_ringparam,
1609 };
1610
1611 static const struct net_device_ops device_ops = {
1612         .ndo_open =                     netvsc_open,
1613         .ndo_stop =                     netvsc_close,
1614         .ndo_start_xmit =               netvsc_start_xmit,
1615         .ndo_change_rx_flags =          netvsc_change_rx_flags,
1616         .ndo_set_rx_mode =              netvsc_set_rx_mode,
1617         .ndo_change_mtu =               netvsc_change_mtu,
1618         .ndo_validate_addr =            eth_validate_addr,
1619         .ndo_set_mac_address =          netvsc_set_mac_addr,
1620         .ndo_select_queue =             netvsc_select_queue,
1621         .ndo_get_stats64 =              netvsc_get_stats64,
1622 #ifdef CONFIG_NET_POLL_CONTROLLER
1623         .ndo_poll_controller =          netvsc_poll_controller,
1624 #endif
1625 };
1626
1627 /*
1628  * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
1629  * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
1630  * present send GARP packet to network peers with netif_notify_peers().
1631  */
1632 static void netvsc_link_change(struct work_struct *w)
1633 {
1634         struct net_device_context *ndev_ctx =
1635                 container_of(w, struct net_device_context, dwork.work);
1636         struct hv_device *device_obj = ndev_ctx->device_ctx;
1637         struct net_device *net = hv_get_drvdata(device_obj);
1638         struct netvsc_device *net_device;
1639         struct rndis_device *rdev;
1640         struct netvsc_reconfig *event = NULL;
1641         bool notify = false, reschedule = false;
1642         unsigned long flags, next_reconfig, delay;
1643
1644         /* if changes are happening, comeback later */
1645         if (!rtnl_trylock()) {
1646                 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1647                 return;
1648         }
1649
1650         net_device = rtnl_dereference(ndev_ctx->nvdev);
1651         if (!net_device)
1652                 goto out_unlock;
1653
1654         rdev = net_device->extension;
1655
1656         next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
1657         if (time_is_after_jiffies(next_reconfig)) {
1658                 /* link_watch only sends one notification with current state
1659                  * per second, avoid doing reconfig more frequently. Handle
1660                  * wrap around.
1661                  */
1662                 delay = next_reconfig - jiffies;
1663                 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
1664                 schedule_delayed_work(&ndev_ctx->dwork, delay);
1665                 goto out_unlock;
1666         }
1667         ndev_ctx->last_reconfig = jiffies;
1668
1669         spin_lock_irqsave(&ndev_ctx->lock, flags);
1670         if (!list_empty(&ndev_ctx->reconfig_events)) {
1671                 event = list_first_entry(&ndev_ctx->reconfig_events,
1672                                          struct netvsc_reconfig, list);
1673                 list_del(&event->list);
1674                 reschedule = !list_empty(&ndev_ctx->reconfig_events);
1675         }
1676         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1677
1678         if (!event)
1679                 goto out_unlock;
1680
1681         switch (event->event) {
1682                 /* Only the following events are possible due to the check in
1683                  * netvsc_linkstatus_callback()
1684                  */
1685         case RNDIS_STATUS_MEDIA_CONNECT:
1686                 if (rdev->link_state) {
1687                         rdev->link_state = false;
1688                         netif_carrier_on(net);
1689                         netif_tx_wake_all_queues(net);
1690                 } else {
1691                         notify = true;
1692                 }
1693                 kfree(event);
1694                 break;
1695         case RNDIS_STATUS_MEDIA_DISCONNECT:
1696                 if (!rdev->link_state) {
1697                         rdev->link_state = true;
1698                         netif_carrier_off(net);
1699                         netif_tx_stop_all_queues(net);
1700                 }
1701                 kfree(event);
1702                 break;
1703         case RNDIS_STATUS_NETWORK_CHANGE:
1704                 /* Only makes sense if carrier is present */
1705                 if (!rdev->link_state) {
1706                         rdev->link_state = true;
1707                         netif_carrier_off(net);
1708                         netif_tx_stop_all_queues(net);
1709                         event->event = RNDIS_STATUS_MEDIA_CONNECT;
1710                         spin_lock_irqsave(&ndev_ctx->lock, flags);
1711                         list_add(&event->list, &ndev_ctx->reconfig_events);
1712                         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1713                         reschedule = true;
1714                 }
1715                 break;
1716         }
1717
1718         rtnl_unlock();
1719
1720         if (notify)
1721                 netdev_notify_peers(net);
1722
1723         /* link_watch only sends one notification with current state per
1724          * second, handle next reconfig event in 2 seconds.
1725          */
1726         if (reschedule)
1727                 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1728
1729         return;
1730
1731 out_unlock:
1732         rtnl_unlock();
1733 }
1734
1735 static struct net_device *get_netvsc_bymac(const u8 *mac)
1736 {
1737         struct net_device *dev;
1738
1739         ASSERT_RTNL();
1740
1741         for_each_netdev(&init_net, dev) {
1742                 if (dev->netdev_ops != &device_ops)
1743                         continue;       /* not a netvsc device */
1744
1745                 if (ether_addr_equal(mac, dev->perm_addr))
1746                         return dev;
1747         }
1748
1749         return NULL;
1750 }
1751
1752 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
1753 {
1754         struct net_device *dev;
1755
1756         ASSERT_RTNL();
1757
1758         for_each_netdev(&init_net, dev) {
1759                 struct net_device_context *net_device_ctx;
1760
1761                 if (dev->netdev_ops != &device_ops)
1762                         continue;       /* not a netvsc device */
1763
1764                 net_device_ctx = netdev_priv(dev);
1765                 if (!rtnl_dereference(net_device_ctx->nvdev))
1766                         continue;       /* device is removed */
1767
1768                 if (rtnl_dereference(net_device_ctx->vf_netdev) == vf_netdev)
1769                         return dev;     /* a match */
1770         }
1771
1772         return NULL;
1773 }
1774
1775 /* Called when VF is injecting data into network stack.
1776  * Change the associated network device from VF to netvsc.
1777  * note: already called with rcu_read_lock
1778  */
1779 static rx_handler_result_t netvsc_vf_handle_frame(struct sk_buff **pskb)
1780 {
1781         struct sk_buff *skb = *pskb;
1782         struct net_device *ndev = rcu_dereference(skb->dev->rx_handler_data);
1783         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1784         struct netvsc_vf_pcpu_stats *pcpu_stats
1785                  = this_cpu_ptr(ndev_ctx->vf_stats);
1786
1787         skb->dev = ndev;
1788
1789         u64_stats_update_begin(&pcpu_stats->syncp);
1790         pcpu_stats->rx_packets++;
1791         pcpu_stats->rx_bytes += skb->len;
1792         u64_stats_update_end(&pcpu_stats->syncp);
1793
1794         return RX_HANDLER_ANOTHER;
1795 }
1796
1797 static int netvsc_vf_join(struct net_device *vf_netdev,
1798                           struct net_device *ndev)
1799 {
1800         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1801         int ret;
1802
1803         ret = netdev_rx_handler_register(vf_netdev,
1804                                          netvsc_vf_handle_frame, ndev);
1805         if (ret != 0) {
1806                 netdev_err(vf_netdev,
1807                            "can not register netvsc VF receive handler (err = %d)\n",
1808                            ret);
1809                 goto rx_handler_failed;
1810         }
1811
1812         ret = netdev_upper_dev_link(vf_netdev, ndev, NULL);
1813         if (ret != 0) {
1814                 netdev_err(vf_netdev,
1815                            "can not set master device %s (err = %d)\n",
1816                            ndev->name, ret);
1817                 goto upper_link_failed;
1818         }
1819
1820         /* set slave flag before open to prevent IPv6 addrconf */
1821         vf_netdev->flags |= IFF_SLAVE;
1822
1823         schedule_delayed_work(&ndev_ctx->vf_takeover, VF_TAKEOVER_INT);
1824
1825         call_netdevice_notifiers(NETDEV_JOIN, vf_netdev);
1826
1827         netdev_info(vf_netdev, "joined to %s\n", ndev->name);
1828         return 0;
1829
1830 upper_link_failed:
1831         netdev_rx_handler_unregister(vf_netdev);
1832 rx_handler_failed:
1833         return ret;
1834 }
1835
1836 static void __netvsc_vf_setup(struct net_device *ndev,
1837                               struct net_device *vf_netdev)
1838 {
1839         int ret;
1840
1841         /* Align MTU of VF with master */
1842         ret = dev_set_mtu(vf_netdev, ndev->mtu);
1843         if (ret)
1844                 netdev_warn(vf_netdev,
1845                             "unable to change mtu to %u\n", ndev->mtu);
1846
1847         /* set multicast etc flags on VF */
1848         dev_change_flags(vf_netdev, ndev->flags | IFF_SLAVE);
1849         dev_uc_sync(vf_netdev, ndev);
1850         dev_mc_sync(vf_netdev, ndev);
1851
1852         if (netif_running(ndev)) {
1853                 ret = dev_open(vf_netdev);
1854                 if (ret)
1855                         netdev_warn(vf_netdev,
1856                                     "unable to open: %d\n", ret);
1857         }
1858 }
1859
1860 /* Setup VF as slave of the synthetic device.
1861  * Runs in workqueue to avoid recursion in netlink callbacks.
1862  */
1863 static void netvsc_vf_setup(struct work_struct *w)
1864 {
1865         struct net_device_context *ndev_ctx
1866                 = container_of(w, struct net_device_context, vf_takeover.work);
1867         struct net_device *ndev = hv_get_drvdata(ndev_ctx->device_ctx);
1868         struct net_device *vf_netdev;
1869
1870         if (!rtnl_trylock()) {
1871                 schedule_delayed_work(&ndev_ctx->vf_takeover, 0);
1872                 return;
1873         }
1874
1875         vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
1876         if (vf_netdev)
1877                 __netvsc_vf_setup(ndev, vf_netdev);
1878
1879         rtnl_unlock();
1880 }
1881
1882 static int netvsc_register_vf(struct net_device *vf_netdev)
1883 {
1884         struct net_device *ndev;
1885         struct net_device_context *net_device_ctx;
1886         struct netvsc_device *netvsc_dev;
1887
1888         if (vf_netdev->addr_len != ETH_ALEN)
1889                 return NOTIFY_DONE;
1890
1891         /*
1892          * We will use the MAC address to locate the synthetic interface to
1893          * associate with the VF interface. If we don't find a matching
1894          * synthetic interface, move on.
1895          */
1896         ndev = get_netvsc_bymac(vf_netdev->perm_addr);
1897         if (!ndev)
1898                 return NOTIFY_DONE;
1899
1900         net_device_ctx = netdev_priv(ndev);
1901         netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
1902         if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
1903                 return NOTIFY_DONE;
1904
1905         if (netvsc_vf_join(vf_netdev, ndev) != 0)
1906                 return NOTIFY_DONE;
1907
1908         netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
1909
1910         dev_hold(vf_netdev);
1911         rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
1912         return NOTIFY_OK;
1913 }
1914
1915 /* VF up/down change detected, schedule to change data path */
1916 static int netvsc_vf_changed(struct net_device *vf_netdev)
1917 {
1918         struct net_device_context *net_device_ctx;
1919         struct netvsc_device *netvsc_dev;
1920         struct net_device *ndev;
1921         bool vf_is_up = netif_running(vf_netdev);
1922
1923         ndev = get_netvsc_byref(vf_netdev);
1924         if (!ndev)
1925                 return NOTIFY_DONE;
1926
1927         net_device_ctx = netdev_priv(ndev);
1928         netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
1929         if (!netvsc_dev)
1930                 return NOTIFY_DONE;
1931
1932         netvsc_switch_datapath(ndev, vf_is_up);
1933         netdev_info(ndev, "Data path switched %s VF: %s\n",
1934                     vf_is_up ? "to" : "from", vf_netdev->name);
1935
1936         return NOTIFY_OK;
1937 }
1938
1939 static int netvsc_unregister_vf(struct net_device *vf_netdev)
1940 {
1941         struct net_device *ndev;
1942         struct net_device_context *net_device_ctx;
1943
1944         ndev = get_netvsc_byref(vf_netdev);
1945         if (!ndev)
1946                 return NOTIFY_DONE;
1947
1948         net_device_ctx = netdev_priv(ndev);
1949         cancel_delayed_work_sync(&net_device_ctx->vf_takeover);
1950
1951         netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
1952
1953         netdev_rx_handler_unregister(vf_netdev);
1954         netdev_upper_dev_unlink(vf_netdev, ndev);
1955         RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
1956         dev_put(vf_netdev);
1957
1958         return NOTIFY_OK;
1959 }
1960
1961 static int netvsc_probe(struct hv_device *dev,
1962                         const struct hv_vmbus_device_id *dev_id)
1963 {
1964         struct net_device *net = NULL;
1965         struct net_device_context *net_device_ctx;
1966         struct netvsc_device_info device_info;
1967         struct netvsc_device *nvdev;
1968         int ret = -ENOMEM;
1969
1970         net = alloc_etherdev_mq(sizeof(struct net_device_context),
1971                                 VRSS_CHANNEL_MAX);
1972         if (!net)
1973                 goto no_net;
1974
1975         netif_carrier_off(net);
1976
1977         netvsc_init_settings(net);
1978
1979         net_device_ctx = netdev_priv(net);
1980         net_device_ctx->device_ctx = dev;
1981         net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
1982         if (netif_msg_probe(net_device_ctx))
1983                 netdev_dbg(net, "netvsc msg_enable: %d\n",
1984                            net_device_ctx->msg_enable);
1985
1986         hv_set_drvdata(dev, net);
1987
1988         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
1989
1990         spin_lock_init(&net_device_ctx->lock);
1991         INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
1992         INIT_DELAYED_WORK(&net_device_ctx->vf_takeover, netvsc_vf_setup);
1993
1994         net_device_ctx->vf_stats
1995                 = netdev_alloc_pcpu_stats(struct netvsc_vf_pcpu_stats);
1996         if (!net_device_ctx->vf_stats)
1997                 goto no_stats;
1998
1999         net->netdev_ops = &device_ops;
2000         net->ethtool_ops = &ethtool_ops;
2001         SET_NETDEV_DEV(net, &dev->device);
2002
2003         /* We always need headroom for rndis header */
2004         net->needed_headroom = RNDIS_AND_PPI_SIZE;
2005
2006         /* Initialize the number of queues to be 1, we may change it if more
2007          * channels are offered later.
2008          */
2009         netif_set_real_num_tx_queues(net, 1);
2010         netif_set_real_num_rx_queues(net, 1);
2011
2012         /* Notify the netvsc driver of the new device */
2013         memset(&device_info, 0, sizeof(device_info));
2014         device_info.num_chn = VRSS_CHANNEL_DEFAULT;
2015         device_info.send_sections = NETVSC_DEFAULT_TX;
2016         device_info.send_section_size = NETVSC_SEND_SECTION_SIZE;
2017         device_info.recv_sections = NETVSC_DEFAULT_RX;
2018         device_info.recv_section_size = NETVSC_RECV_SECTION_SIZE;
2019
2020         nvdev = rndis_filter_device_add(dev, &device_info);
2021         if (IS_ERR(nvdev)) {
2022                 ret = PTR_ERR(nvdev);
2023                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
2024                 goto rndis_failed;
2025         }
2026
2027         memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
2028
2029         /* hw_features computed in rndis_netdev_set_hwcaps() */
2030         net->features = net->hw_features |
2031                 NETIF_F_HIGHDMA | NETIF_F_SG |
2032                 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
2033         net->vlan_features = net->features;
2034
2035         netdev_lockdep_set_classes(net);
2036
2037         /* MTU range: 68 - 1500 or 65521 */
2038         net->min_mtu = NETVSC_MTU_MIN;
2039         if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
2040                 net->max_mtu = NETVSC_MTU - ETH_HLEN;
2041         else
2042                 net->max_mtu = ETH_DATA_LEN;
2043
2044         ret = register_netdev(net);
2045         if (ret != 0) {
2046                 pr_err("Unable to register netdev.\n");
2047                 goto register_failed;
2048         }
2049
2050         return ret;
2051
2052 register_failed:
2053         rndis_filter_device_remove(dev, nvdev);
2054 rndis_failed:
2055         free_percpu(net_device_ctx->vf_stats);
2056 no_stats:
2057         hv_set_drvdata(dev, NULL);
2058         free_netdev(net);
2059 no_net:
2060         return ret;
2061 }
2062
2063 static int netvsc_remove(struct hv_device *dev)
2064 {
2065         struct net_device_context *ndev_ctx;
2066         struct net_device *vf_netdev;
2067         struct net_device *net;
2068
2069         net = hv_get_drvdata(dev);
2070         if (net == NULL) {
2071                 dev_err(&dev->device, "No net device to remove\n");
2072                 return 0;
2073         }
2074
2075         ndev_ctx = netdev_priv(net);
2076
2077         netif_device_detach(net);
2078
2079         cancel_delayed_work_sync(&ndev_ctx->dwork);
2080
2081         /*
2082          * Call to the vsc driver to let it know that the device is being
2083          * removed. Also blocks mtu and channel changes.
2084          */
2085         rtnl_lock();
2086         vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2087         if (vf_netdev)
2088                 netvsc_unregister_vf(vf_netdev);
2089
2090         unregister_netdevice(net);
2091
2092         rndis_filter_device_remove(dev,
2093                                    rtnl_dereference(ndev_ctx->nvdev));
2094         rtnl_unlock();
2095
2096         hv_set_drvdata(dev, NULL);
2097
2098         free_percpu(ndev_ctx->vf_stats);
2099         free_netdev(net);
2100         return 0;
2101 }
2102
2103 static const struct hv_vmbus_device_id id_table[] = {
2104         /* Network guid */
2105         { HV_NIC_GUID, },
2106         { },
2107 };
2108
2109 MODULE_DEVICE_TABLE(vmbus, id_table);
2110
2111 /* The one and only one */
2112 static struct  hv_driver netvsc_drv = {
2113         .name = KBUILD_MODNAME,
2114         .id_table = id_table,
2115         .probe = netvsc_probe,
2116         .remove = netvsc_remove,
2117 };
2118
2119 /*
2120  * On Hyper-V, every VF interface is matched with a corresponding
2121  * synthetic interface. The synthetic interface is presented first
2122  * to the guest. When the corresponding VF instance is registered,
2123  * we will take care of switching the data path.
2124  */
2125 static int netvsc_netdev_event(struct notifier_block *this,
2126                                unsigned long event, void *ptr)
2127 {
2128         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2129
2130         /* Skip our own events */
2131         if (event_dev->netdev_ops == &device_ops)
2132                 return NOTIFY_DONE;
2133
2134         /* Avoid non-Ethernet type devices */
2135         if (event_dev->type != ARPHRD_ETHER)
2136                 return NOTIFY_DONE;
2137
2138         /* Avoid Vlan dev with same MAC registering as VF */
2139         if (is_vlan_dev(event_dev))
2140                 return NOTIFY_DONE;
2141
2142         /* Avoid Bonding master dev with same MAC registering as VF */
2143         if ((event_dev->priv_flags & IFF_BONDING) &&
2144             (event_dev->flags & IFF_MASTER))
2145                 return NOTIFY_DONE;
2146
2147         switch (event) {
2148         case NETDEV_REGISTER:
2149                 return netvsc_register_vf(event_dev);
2150         case NETDEV_UNREGISTER:
2151                 return netvsc_unregister_vf(event_dev);
2152         case NETDEV_UP:
2153         case NETDEV_DOWN:
2154                 return netvsc_vf_changed(event_dev);
2155         default:
2156                 return NOTIFY_DONE;
2157         }
2158 }
2159
2160 static struct notifier_block netvsc_netdev_notifier = {
2161         .notifier_call = netvsc_netdev_event,
2162 };
2163
2164 static void __exit netvsc_drv_exit(void)
2165 {
2166         unregister_netdevice_notifier(&netvsc_netdev_notifier);
2167         vmbus_driver_unregister(&netvsc_drv);
2168 }
2169
2170 static int __init netvsc_drv_init(void)
2171 {
2172         int ret;
2173
2174         if (ring_size < RING_SIZE_MIN) {
2175                 ring_size = RING_SIZE_MIN;
2176                 pr_info("Increased ring_size to %u (min allowed)\n",
2177                         ring_size);
2178         }
2179         netvsc_ring_bytes = ring_size * PAGE_SIZE;
2180         netvsc_ring_reciprocal = reciprocal_value(netvsc_ring_bytes);
2181
2182         ret = vmbus_driver_register(&netvsc_drv);
2183         if (ret)
2184                 return ret;
2185
2186         register_netdevice_notifier(&netvsc_netdev_notifier);
2187         return 0;
2188 }
2189
2190 MODULE_LICENSE("GPL");
2191 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
2192
2193 module_init(netvsc_drv_init);
2194 module_exit(netvsc_drv_exit);