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/pci.h>
  33 #include <linux/skbuff.h>
  34 #include <linux/if_vlan.h>
  35 #include <linux/in.h>
  36 #include <linux/slab.h>
  37 #include <linux/rtnetlink.h>
  38 #include <linux/netpoll.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 #define RETRY_US_LO     5000
  51 #define RETRY_US_HI     10000
  52 #define RETRY_MAX       2000    /* >10 sec */
  53
  54 #define LINKCHANGE_INT (2 * HZ)
  55 #define VF_TAKEOVER_INT (HZ / 10)
  56
  57 static unsigned int ring_size __ro_after_init = 128;
  58 module_param(ring_size, uint, 0444);
  59 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
  60 unsigned int netvsc_ring_bytes __ro_after_init;
  61
  62 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
  63                                 NETIF_MSG_LINK | NETIF_MSG_IFUP |
  64                                 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
  65                                 NETIF_MSG_TX_ERR;
  66
  67 static int debug = -1;
  68 module_param(debug, int, 0444);
  69 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  70
  71 static LIST_HEAD(netvsc_dev_list);
  72
  73 static void netvsc_change_rx_flags(struct net_device *net, int change)
  74 {
  75         struct net_device_context *ndev_ctx = netdev_priv(net);
  76         struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
  77         int inc;
  78
  79         if (!vf_netdev)
  80                 return;
  81
  82         if (change & IFF_PROMISC) {
  83                 inc = (net->flags & IFF_PROMISC) ? 1 : -1;
  84                 dev_set_promiscuity(vf_netdev, inc);
  85         }
  86
  87         if (change & IFF_ALLMULTI) {
  88                 inc = (net->flags & IFF_ALLMULTI) ? 1 : -1;
  89                 dev_set_allmulti(vf_netdev, inc);
  90         }
  91 }
  92
  93 static void netvsc_set_rx_mode(struct net_device *net)
  94 {
  95         struct net_device_context *ndev_ctx = netdev_priv(net);
  96         struct net_device *vf_netdev;
  97         struct netvsc_device *nvdev;
  98
  99         rcu_read_lock();
 100         vf_netdev = rcu_dereference(ndev_ctx->vf_netdev);
 101         if (vf_netdev) {
 102                 dev_uc_sync(vf_netdev, net);
 103                 dev_mc_sync(vf_netdev, net);
 104         }
 105
 106         nvdev = rcu_dereference(ndev_ctx->nvdev);
 107         if (nvdev)
 108                 rndis_filter_update(nvdev);
 109         rcu_read_unlock();
 110 }
 111
 112 static int netvsc_open(struct net_device *net)
 113 {
 114         struct net_device_context *ndev_ctx = netdev_priv(net);
 115         struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
 116         struct netvsc_device *nvdev = rtnl_dereference(ndev_ctx->nvdev);
 117         struct rndis_device *rdev;
 118         int ret = 0;
 119
 120         netif_carrier_off(net);
 121
 122         /* Open up the device */
 123         ret = rndis_filter_open(nvdev);
 124         if (ret != 0) {
 125                 netdev_err(net, "unable to open device (ret %d).\n", ret);
 126                 return ret;
 127         }
 128
 129         rdev = nvdev->extension;
 130         if (!rdev->link_state) {
 131                 netif_carrier_on(net);
 132                 netif_tx_wake_all_queues(net);
 133         }
 134
 135         if (vf_netdev) {
 136                 /* Setting synthetic device up transparently sets
 137                  * slave as up. If open fails, then slave will be
 138                  * still be offline (and not used).
 139                  */
 140                 ret = dev_open(vf_netdev);
 141                 if (ret)
 142                         netdev_warn(net,
 143                                     "unable to open slave: %s: %d\n",
 144                                     vf_netdev->name, ret);
 145         }
 146         return 0;
 147 }
 148
 149 static int netvsc_wait_until_empty(struct netvsc_device *nvdev)
 150 {
 151         unsigned int retry = 0;
 152         int i;
 153
 154         /* Ensure pending bytes in ring are read */
 155         for (;;) {
 156                 u32 aread = 0;
 157
 158                 for (i = 0; i < nvdev->num_chn; i++) {
 159                         struct vmbus_channel *chn
 160                                 = nvdev->chan_table[i].channel;
 161
 162                         if (!chn)
 163                                 continue;
 164
 165                         /* make sure receive not running now */
 166                         napi_synchronize(&nvdev->chan_table[i].napi);
 167
 168                         aread = hv_get_bytes_to_read(&chn->inbound);
 169                         if (aread)
 170                                 break;
 171
 172                         aread = hv_get_bytes_to_read(&chn->outbound);
 173                         if (aread)
 174                                 break;
 175                 }
 176
 177                 if (aread == 0)
 178                         return 0;
 179
 180                 if (++retry > RETRY_MAX)
 181                         return -ETIMEDOUT;
 182
 183                 usleep_range(RETRY_US_LO, RETRY_US_HI);
 184         }
 185 }
 186
 187 static int netvsc_close(struct net_device *net)
 188 {
 189         struct net_device_context *net_device_ctx = netdev_priv(net);
 190         struct net_device *vf_netdev
 191                 = rtnl_dereference(net_device_ctx->vf_netdev);
 192         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
 193         int ret;
 194
 195         netif_tx_disable(net);
 196
 197         /* No need to close rndis filter if it is removed already */
 198         if (!nvdev)
 199                 return 0;
 200
 201         ret = rndis_filter_close(nvdev);
 202         if (ret != 0) {
 203                 netdev_err(net, "unable to close device (ret %d).\n", ret);
 204                 return ret;
 205         }
 206
 207         ret = netvsc_wait_until_empty(nvdev);
 208         if (ret)
 209                 netdev_err(net, "Ring buffer not empty after closing rndis\n");
 210
 211         if (vf_netdev)
 212                 dev_close(vf_netdev);
 213
 214         return ret;
 215 }
 216
 217 static inline void *init_ppi_data(struct rndis_message *msg,
 218                                   u32 ppi_size, u32 pkt_type)
 219 {
 220         struct rndis_packet *rndis_pkt = &msg->msg.pkt;
 221         struct rndis_per_packet_info *ppi;
 222
 223         rndis_pkt->data_offset += ppi_size;
 224         ppi = (void *)rndis_pkt + rndis_pkt->per_pkt_info_offset
 225                 + rndis_pkt->per_pkt_info_len;
 226
 227         ppi->size = ppi_size;
 228         ppi->type = pkt_type;
 229         ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
 230
 231         rndis_pkt->per_pkt_info_len += ppi_size;
 232
 233         return ppi + 1;
 234 }
 235
 236 /* Azure hosts don't support non-TCP port numbers in hashing for fragmented
 237  * packets. We can use ethtool to change UDP hash level when necessary.
 238  */
 239 static inline u32 netvsc_get_hash(
 240         struct sk_buff *skb,
 241         const struct net_device_context *ndc)
 242 {
 243         struct flow_keys flow;
 244         u32 hash, pkt_proto = 0;
 245         static u32 hashrnd __read_mostly;
 246
 247         net_get_random_once(&hashrnd, sizeof(hashrnd));
 248
 249         if (!skb_flow_dissect_flow_keys(skb, &flow, 0))
 250                 return 0;
 251
 252         switch (flow.basic.ip_proto) {
 253         case IPPROTO_TCP:
 254                 if (flow.basic.n_proto == htons(ETH_P_IP))
 255                         pkt_proto = HV_TCP4_L4HASH;
 256                 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
 257                         pkt_proto = HV_TCP6_L4HASH;
 258
 259                 break;
 260
 261         case IPPROTO_UDP:
 262                 if (flow.basic.n_proto == htons(ETH_P_IP))
 263                         pkt_proto = HV_UDP4_L4HASH;
 264                 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
 265                         pkt_proto = HV_UDP6_L4HASH;
 266
 267                 break;
 268         }
 269
 270         if (pkt_proto & ndc->l4_hash) {
 271                 return skb_get_hash(skb);
 272         } else {
 273                 if (flow.basic.n_proto == htons(ETH_P_IP))
 274                         hash = jhash2((u32 *)&flow.addrs.v4addrs, 2, hashrnd);
 275                 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
 276                         hash = jhash2((u32 *)&flow.addrs.v6addrs, 8, hashrnd);
 277                 else
 278                         hash = 0;
 279
 280                 skb_set_hash(skb, hash, PKT_HASH_TYPE_L3);
 281         }
 282
 283         return hash;
 284 }
 285
 286 static inline int netvsc_get_tx_queue(struct net_device *ndev,
 287                                       struct sk_buff *skb, int old_idx)
 288 {
 289         const struct net_device_context *ndc = netdev_priv(ndev);
 290         struct sock *sk = skb->sk;
 291         int q_idx;
 292
 293         q_idx = ndc->tx_table[netvsc_get_hash(skb, ndc) &
 294                               (VRSS_SEND_TAB_SIZE - 1)];
 295
 296         /* If queue index changed record the new value */
 297         if (q_idx != old_idx &&
 298             sk && sk_fullsock(sk) && rcu_access_pointer(sk->sk_dst_cache))
 299                 sk_tx_queue_set(sk, q_idx);
 300
 301         return q_idx;
 302 }
 303
 304 /*
 305  * Select queue for transmit.
 306  *
 307  * If a valid queue has already been assigned, then use that.
 308  * Otherwise compute tx queue based on hash and the send table.
 309  *
 310  * This is basically similar to default (__netdev_pick_tx) with the added step
 311  * of using the host send_table when no other queue has been assigned.
 312  *
 313  * TODO support XPS - but get_xps_queue not exported
 314  */
 315 static u16 netvsc_pick_tx(struct net_device *ndev, struct sk_buff *skb)
 316 {
 317         int q_idx = sk_tx_queue_get(skb->sk);
 318
 319         if (q_idx < 0 || skb->ooo_okay || q_idx >= ndev->real_num_tx_queues) {
 320                 /* If forwarding a packet, we use the recorded queue when
 321                  * available for better cache locality.
 322                  */
 323                 if (skb_rx_queue_recorded(skb))
 324                         q_idx = skb_get_rx_queue(skb);
 325                 else
 326                         q_idx = netvsc_get_tx_queue(ndev, skb, q_idx);
 327         }
 328
 329         return q_idx;
 330 }
 331
 332 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
 333                                struct net_device *sb_dev,
 334                                select_queue_fallback_t fallback)
 335 {
 336         struct net_device_context *ndc = netdev_priv(ndev);
 337         struct net_device *vf_netdev;
 338         u16 txq;
 339
 340         rcu_read_lock();
 341         vf_netdev = rcu_dereference(ndc->vf_netdev);
 342         if (vf_netdev) {
 343                 const struct net_device_ops *vf_ops = vf_netdev->netdev_ops;
 344
 345                 if (vf_ops->ndo_select_queue)
 346                         txq = vf_ops->ndo_select_queue(vf_netdev, skb,
 347                                                        sb_dev, fallback);
 348                 else
 349                         txq = fallback(vf_netdev, skb, NULL);
 350
 351                 /* Record the queue selected by VF so that it can be
 352                  * used for common case where VF has more queues than
 353                  * the synthetic device.
 354                  */
 355                 qdisc_skb_cb(skb)->slave_dev_queue_mapping = txq;
 356         } else {
 357                 txq = netvsc_pick_tx(ndev, skb);
 358         }
 359         rcu_read_unlock();
 360
 361         while (unlikely(txq >= ndev->real_num_tx_queues))
 362                 txq -= ndev->real_num_tx_queues;
 363
 364         return txq;
 365 }
 366
 367 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
 368                        struct hv_page_buffer *pb)
 369 {
 370         int j = 0;
 371
 372         /* Deal with compund pages by ignoring unused part
 373          * of the page.
 374          */
 375         page += (offset >> PAGE_SHIFT);
 376         offset &= ~PAGE_MASK;
 377
 378         while (len > 0) {
 379                 unsigned long bytes;
 380
 381                 bytes = PAGE_SIZE - offset;
 382                 if (bytes > len)
 383                         bytes = len;
 384                 pb[j].pfn = page_to_pfn(page);
 385                 pb[j].offset = offset;
 386                 pb[j].len = bytes;
 387
 388                 offset += bytes;
 389                 len -= bytes;
 390
 391                 if (offset == PAGE_SIZE && len) {
 392                         page++;
 393                         offset = 0;
 394                         j++;
 395                 }
 396         }
 397
 398         return j + 1;
 399 }
 400
 401 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
 402                            struct hv_netvsc_packet *packet,
 403                            struct hv_page_buffer *pb)
 404 {
 405         u32 slots_used = 0;
 406         char *data = skb->data;
 407         int frags = skb_shinfo(skb)->nr_frags;
 408         int i;
 409
 410         /* The packet is laid out thus:
 411          * 1. hdr: RNDIS header and PPI
 412          * 2. skb linear data
 413          * 3. skb fragment data
 414          */
 415         slots_used += fill_pg_buf(virt_to_page(hdr),
 416                                   offset_in_page(hdr),
 417                                   len, &pb[slots_used]);
 418
 419         packet->rmsg_size = len;
 420         packet->rmsg_pgcnt = slots_used;
 421
 422         slots_used += fill_pg_buf(virt_to_page(data),
 423                                 offset_in_page(data),
 424                                 skb_headlen(skb), &pb[slots_used]);
 425
 426         for (i = 0; i < frags; i++) {
 427                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 428
 429                 slots_used += fill_pg_buf(skb_frag_page(frag),
 430                                         frag->page_offset,
 431                                         skb_frag_size(frag), &pb[slots_used]);
 432         }
 433         return slots_used;
 434 }
 435
 436 static int count_skb_frag_slots(struct sk_buff *skb)
 437 {
 438         int i, frags = skb_shinfo(skb)->nr_frags;
 439         int pages = 0;
 440
 441         for (i = 0; i < frags; i++) {
 442                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 443                 unsigned long size = skb_frag_size(frag);
 444                 unsigned long offset = frag->page_offset;
 445
 446                 /* Skip unused frames from start of page */
 447                 offset &= ~PAGE_MASK;
 448                 pages += PFN_UP(offset + size);
 449         }
 450         return pages;
 451 }
 452
 453 static int netvsc_get_slots(struct sk_buff *skb)
 454 {
 455         char *data = skb->data;
 456         unsigned int offset = offset_in_page(data);
 457         unsigned int len = skb_headlen(skb);
 458         int slots;
 459         int frag_slots;
 460
 461         slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
 462         frag_slots = count_skb_frag_slots(skb);
 463         return slots + frag_slots;
 464 }
 465
 466 static u32 net_checksum_info(struct sk_buff *skb)
 467 {
 468         if (skb->protocol == htons(ETH_P_IP)) {
 469                 struct iphdr *ip = ip_hdr(skb);
 470
 471                 if (ip->protocol == IPPROTO_TCP)
 472                         return TRANSPORT_INFO_IPV4_TCP;
 473                 else if (ip->protocol == IPPROTO_UDP)
 474                         return TRANSPORT_INFO_IPV4_UDP;
 475         } else {
 476                 struct ipv6hdr *ip6 = ipv6_hdr(skb);
 477
 478                 if (ip6->nexthdr == IPPROTO_TCP)
 479                         return TRANSPORT_INFO_IPV6_TCP;
 480                 else if (ip6->nexthdr == IPPROTO_UDP)
 481                         return TRANSPORT_INFO_IPV6_UDP;
 482         }
 483
 484         return TRANSPORT_INFO_NOT_IP;
 485 }
 486
 487 /* Send skb on the slave VF device. */
 488 static int netvsc_vf_xmit(struct net_device *net, struct net_device *vf_netdev,
 489                           struct sk_buff *skb)
 490 {
 491         struct net_device_context *ndev_ctx = netdev_priv(net);
 492         unsigned int len = skb->len;
 493         int rc;
 494
 495         skb->dev = vf_netdev;
 496         skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
 497
 498         rc = dev_queue_xmit(skb);
 499         if (likely(rc == NET_XMIT_SUCCESS || rc == NET_XMIT_CN)) {
 500                 struct netvsc_vf_pcpu_stats *pcpu_stats
 501                         = this_cpu_ptr(ndev_ctx->vf_stats);
 502
 503                 u64_stats_update_begin(&pcpu_stats->syncp);
 504                 pcpu_stats->tx_packets++;
 505                 pcpu_stats->tx_bytes += len;
 506                 u64_stats_update_end(&pcpu_stats->syncp);
 507         } else {
 508                 this_cpu_inc(ndev_ctx->vf_stats->tx_dropped);
 509         }
 510
 511         return rc;
 512 }
 513
 514 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
 515 {
 516         struct net_device_context *net_device_ctx = netdev_priv(net);
 517         struct hv_netvsc_packet *packet = NULL;
 518         int ret;
 519         unsigned int num_data_pgs;
 520         struct rndis_message *rndis_msg;
 521         struct net_device *vf_netdev;
 522         u32 rndis_msg_size;
 523         u32 hash;
 524         struct hv_page_buffer pb[MAX_PAGE_BUFFER_COUNT];
 525
 526         /* if VF is present and up then redirect packets
 527          * already called with rcu_read_lock_bh
 528          */
 529         vf_netdev = rcu_dereference_bh(net_device_ctx->vf_netdev);
 530         if (vf_netdev && netif_running(vf_netdev) &&
 531             !netpoll_tx_running(net))
 532                 return netvsc_vf_xmit(net, vf_netdev, skb);
 533
 534         /* We will atmost need two pages to describe the rndis
 535          * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
 536          * of pages in a single packet. If skb is scattered around
 537          * more pages we try linearizing it.
 538          */
 539
 540         num_data_pgs = netvsc_get_slots(skb) + 2;
 541
 542         if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
 543                 ++net_device_ctx->eth_stats.tx_scattered;
 544
 545                 if (skb_linearize(skb))
 546                         goto no_memory;
 547
 548                 num_data_pgs = netvsc_get_slots(skb) + 2;
 549                 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
 550                         ++net_device_ctx->eth_stats.tx_too_big;
 551                         goto drop;
 552                 }
 553         }
 554
 555         /*
 556          * Place the rndis header in the skb head room and
 557          * the skb->cb will be used for hv_netvsc_packet
 558          * structure.
 559          */
 560         ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
 561         if (ret)
 562                 goto no_memory;
 563
 564         /* Use the skb control buffer for building up the packet */
 565         BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
 566                         FIELD_SIZEOF(struct sk_buff, cb));
 567         packet = (struct hv_netvsc_packet *)skb->cb;
 568
 569         packet->q_idx = skb_get_queue_mapping(skb);
 570
 571         packet->total_data_buflen = skb->len;
 572         packet->total_bytes = skb->len;
 573         packet->total_packets = 1;
 574
 575         rndis_msg = (struct rndis_message *)skb->head;
 576
 577         /* Add the rndis header */
 578         rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
 579         rndis_msg->msg_len = packet->total_data_buflen;
 580
 581         rndis_msg->msg.pkt = (struct rndis_packet) {
 582                 .data_offset = sizeof(struct rndis_packet),
 583                 .data_len = packet->total_data_buflen,
 584                 .per_pkt_info_offset = sizeof(struct rndis_packet),
 585         };
 586
 587         rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
 588
 589         hash = skb_get_hash_raw(skb);
 590         if (hash != 0 && net->real_num_tx_queues > 1) {
 591                 u32 *hash_info;
 592
 593                 rndis_msg_size += NDIS_HASH_PPI_SIZE;
 594                 hash_info = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
 595                                           NBL_HASH_VALUE);
 596                 *hash_info = hash;
 597         }
 598
 599         if (skb_vlan_tag_present(skb)) {
 600                 struct ndis_pkt_8021q_info *vlan;
 601
 602                 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
 603                 vlan = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
 604                                      IEEE_8021Q_INFO);
 605
 606                 vlan->value = 0;
 607                 vlan->vlanid = skb->vlan_tci & VLAN_VID_MASK;
 608                 vlan->pri = (skb->vlan_tci & VLAN_PRIO_MASK) >>
 609                                 VLAN_PRIO_SHIFT;
 610         }
 611
 612         if (skb_is_gso(skb)) {
 613                 struct ndis_tcp_lso_info *lso_info;
 614
 615                 rndis_msg_size += NDIS_LSO_PPI_SIZE;
 616                 lso_info = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
 617                                          TCP_LARGESEND_PKTINFO);
 618
 619                 lso_info->value = 0;
 620                 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
 621                 if (skb->protocol == htons(ETH_P_IP)) {
 622                         lso_info->lso_v2_transmit.ip_version =
 623                                 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
 624                         ip_hdr(skb)->tot_len = 0;
 625                         ip_hdr(skb)->check = 0;
 626                         tcp_hdr(skb)->check =
 627                                 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
 628                                                    ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 629                 } else {
 630                         lso_info->lso_v2_transmit.ip_version =
 631                                 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
 632                         ipv6_hdr(skb)->payload_len = 0;
 633                         tcp_hdr(skb)->check =
 634                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 635                                                  &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 636                 }
 637                 lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb);
 638                 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
 639         } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
 640                 if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) {
 641                         struct ndis_tcp_ip_checksum_info *csum_info;
 642
 643                         rndis_msg_size += NDIS_CSUM_PPI_SIZE;
 644                         csum_info = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
 645                                                   TCPIP_CHKSUM_PKTINFO);
 646
 647                         csum_info->value = 0;
 648                         csum_info->transmit.tcp_header_offset = skb_transport_offset(skb);
 649
 650                         if (skb->protocol == htons(ETH_P_IP)) {
 651                                 csum_info->transmit.is_ipv4 = 1;
 652
 653                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
 654                                         csum_info->transmit.tcp_checksum = 1;
 655                                 else
 656                                         csum_info->transmit.udp_checksum = 1;
 657                         } else {
 658                                 csum_info->transmit.is_ipv6 = 1;
 659
 660                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
 661                                         csum_info->transmit.tcp_checksum = 1;
 662                                 else
 663                                         csum_info->transmit.udp_checksum = 1;
 664                         }
 665                 } else {
 666                         /* Can't do offload of this type of checksum */
 667                         if (skb_checksum_help(skb))
 668                                 goto drop;
 669                 }
 670         }
 671
 672         /* Start filling in the page buffers with the rndis hdr */
 673         rndis_msg->msg_len += rndis_msg_size;
 674         packet->total_data_buflen = rndis_msg->msg_len;
 675         packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
 676                                                skb, packet, pb);
 677
 678         /* timestamp packet in software */
 679         skb_tx_timestamp(skb);
 680
 681         ret = netvsc_send(net, packet, rndis_msg, pb, skb);
 682         if (likely(ret == 0))
 683                 return NETDEV_TX_OK;
 684
 685         if (ret == -EAGAIN) {
 686                 ++net_device_ctx->eth_stats.tx_busy;
 687                 return NETDEV_TX_BUSY;
 688         }
 689
 690         if (ret == -ENOSPC)
 691                 ++net_device_ctx->eth_stats.tx_no_space;
 692
 693 drop:
 694         dev_kfree_skb_any(skb);
 695         net->stats.tx_dropped++;
 696
 697         return NETDEV_TX_OK;
 698
 699 no_memory:
 700         ++net_device_ctx->eth_stats.tx_no_memory;
 701         goto drop;
 702 }
 703
 704 /*
 705  * netvsc_linkstatus_callback - Link up/down notification
 706  */
 707 void netvsc_linkstatus_callback(struct net_device *net,
 708                                 struct rndis_message *resp)
 709 {
 710         struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
 711         struct net_device_context *ndev_ctx = netdev_priv(net);
 712         struct netvsc_reconfig *event;
 713         unsigned long flags;
 714
 715         /* Update the physical link speed when changing to another vSwitch */
 716         if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
 717                 u32 speed;
 718
 719                 speed = *(u32 *)((void *)indicate
 720                                  + indicate->status_buf_offset) / 10000;
 721                 ndev_ctx->speed = speed;
 722                 return;
 723         }
 724
 725         /* Handle these link change statuses below */
 726         if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
 727             indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
 728             indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
 729                 return;
 730
 731         if (net->reg_state != NETREG_REGISTERED)
 732                 return;
 733
 734         event = kzalloc(sizeof(*event), GFP_ATOMIC);
 735         if (!event)
 736                 return;
 737         event->event = indicate->status;
 738
 739         spin_lock_irqsave(&ndev_ctx->lock, flags);
 740         list_add_tail(&event->list, &ndev_ctx->reconfig_events);
 741         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
 742
 743         schedule_delayed_work(&ndev_ctx->dwork, 0);
 744 }
 745
 746 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
 747                                              struct napi_struct *napi,
 748                                              const struct ndis_tcp_ip_checksum_info *csum_info,
 749                                              const struct ndis_pkt_8021q_info *vlan,
 750                                              void *data, u32 buflen)
 751 {
 752         struct sk_buff *skb;
 753
 754         skb = napi_alloc_skb(napi, buflen);
 755         if (!skb)
 756                 return skb;
 757
 758         /*
 759          * Copy to skb. This copy is needed here since the memory pointed by
 760          * hv_netvsc_packet cannot be deallocated
 761          */
 762         skb_put_data(skb, data, buflen);
 763
 764         skb->protocol = eth_type_trans(skb, net);
 765
 766         /* skb is already created with CHECKSUM_NONE */
 767         skb_checksum_none_assert(skb);
 768
 769         /*
 770          * In Linux, the IP checksum is always checked.
 771          * Do L4 checksum offload if enabled and present.
 772          */
 773         if (csum_info && (net->features & NETIF_F_RXCSUM)) {
 774                 if (csum_info->receive.tcp_checksum_succeeded ||
 775                     csum_info->receive.udp_checksum_succeeded)
 776                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 777         }
 778
 779         if (vlan) {
 780                 u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT);
 781
 782                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 783                                        vlan_tci);
 784         }
 785
 786         return skb;
 787 }
 788
 789 /*
 790  * netvsc_recv_callback -  Callback when we receive a packet from the
 791  * "wire" on the specified device.
 792  */
 793 int netvsc_recv_callback(struct net_device *net,
 794                          struct netvsc_device *net_device,
 795                          struct vmbus_channel *channel,
 796                          void  *data, u32 len,
 797                          const struct ndis_tcp_ip_checksum_info *csum_info,
 798                          const struct ndis_pkt_8021q_info *vlan)
 799 {
 800         struct net_device_context *net_device_ctx = netdev_priv(net);
 801         u16 q_idx = channel->offermsg.offer.sub_channel_index;
 802         struct netvsc_channel *nvchan = &net_device->chan_table[q_idx];
 803         struct sk_buff *skb;
 804         struct netvsc_stats *rx_stats;
 805
 806         if (net->reg_state != NETREG_REGISTERED)
 807                 return NVSP_STAT_FAIL;
 808
 809         /* Allocate a skb - TODO direct I/O to pages? */
 810         skb = netvsc_alloc_recv_skb(net, &nvchan->napi,
 811                                     csum_info, vlan, data, len);
 812         if (unlikely(!skb)) {
 813                 ++net_device_ctx->eth_stats.rx_no_memory;
 814                 rcu_read_unlock();
 815                 return NVSP_STAT_FAIL;
 816         }
 817
 818         skb_record_rx_queue(skb, q_idx);
 819
 820         /*
 821          * Even if injecting the packet, record the statistics
 822          * on the synthetic device because modifying the VF device
 823          * statistics will not work correctly.
 824          */
 825         rx_stats = &nvchan->rx_stats;
 826         u64_stats_update_begin(&rx_stats->syncp);
 827         rx_stats->packets++;
 828         rx_stats->bytes += len;
 829
 830         if (skb->pkt_type == PACKET_BROADCAST)
 831                 ++rx_stats->broadcast;
 832         else if (skb->pkt_type == PACKET_MULTICAST)
 833                 ++rx_stats->multicast;
 834         u64_stats_update_end(&rx_stats->syncp);
 835
 836         napi_gro_receive(&nvchan->napi, skb);
 837         return NVSP_STAT_SUCCESS;
 838 }
 839
 840 static void netvsc_get_drvinfo(struct net_device *net,
 841                                struct ethtool_drvinfo *info)
 842 {
 843         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
 844         strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
 845 }
 846
 847 static void netvsc_get_channels(struct net_device *net,
 848                                 struct ethtool_channels *channel)
 849 {
 850         struct net_device_context *net_device_ctx = netdev_priv(net);
 851         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
 852
 853         if (nvdev) {
 854                 channel->max_combined   = nvdev->max_chn;
 855                 channel->combined_count = nvdev->num_chn;
 856         }
 857 }
 858
 859 static int netvsc_detach(struct net_device *ndev,
 860                          struct netvsc_device *nvdev)
 861 {
 862         struct net_device_context *ndev_ctx = netdev_priv(ndev);
 863         struct hv_device *hdev = ndev_ctx->device_ctx;
 864         int ret;
 865
 866         /* Don't try continuing to try and setup sub channels */
 867         if (cancel_work_sync(&nvdev->subchan_work))
 868                 nvdev->num_chn = 1;
 869
 870         /* If device was up (receiving) then shutdown */
 871         if (netif_running(ndev)) {
 872                 netif_tx_disable(ndev);
 873
 874                 ret = rndis_filter_close(nvdev);
 875                 if (ret) {
 876                         netdev_err(ndev,
 877                                    "unable to close device (ret %d).\n", ret);
 878                         return ret;
 879                 }
 880
 881                 ret = netvsc_wait_until_empty(nvdev);
 882                 if (ret) {
 883                         netdev_err(ndev,
 884                                    "Ring buffer not empty after closing rndis\n");
 885                         return ret;
 886                 }
 887         }
 888
 889         netif_device_detach(ndev);
 890
 891         rndis_filter_device_remove(hdev, nvdev);
 892
 893         return 0;
 894 }
 895
 896 static int netvsc_attach(struct net_device *ndev,
 897                          struct netvsc_device_info *dev_info)
 898 {
 899         struct net_device_context *ndev_ctx = netdev_priv(ndev);
 900         struct hv_device *hdev = ndev_ctx->device_ctx;
 901         struct netvsc_device *nvdev;
 902         struct rndis_device *rdev;
 903         int ret;
 904
 905         nvdev = rndis_filter_device_add(hdev, dev_info);
 906         if (IS_ERR(nvdev))
 907                 return PTR_ERR(nvdev);
 908
 909         if (nvdev->num_chn > 1) {
 910                 ret = rndis_set_subchannel(ndev, nvdev);
 911
 912                 /* if unavailable, just proceed with one queue */
 913                 if (ret) {
 914                         nvdev->max_chn = 1;
 915                         nvdev->num_chn = 1;
 916                 }
 917         }
 918
 919         /* In any case device is now ready */
 920         netif_device_attach(ndev);
 921
 922         /* Note: enable and attach happen when sub-channels setup */
 923         netif_carrier_off(ndev);
 924
 925         if (netif_running(ndev)) {
 926                 ret = rndis_filter_open(nvdev);
 927                 if (ret)
 928                         return ret;
 929
 930                 rdev = nvdev->extension;
 931                 if (!rdev->link_state)
 932                         netif_carrier_on(ndev);
 933         }
 934
 935         return 0;
 936 }
 937
 938 static int netvsc_set_channels(struct net_device *net,
 939                                struct ethtool_channels *channels)
 940 {
 941         struct net_device_context *net_device_ctx = netdev_priv(net);
 942         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
 943         unsigned int orig, count = channels->combined_count;
 944         struct netvsc_device_info device_info;
 945         int ret;
 946
 947         /* We do not support separate count for rx, tx, or other */
 948         if (count == 0 ||
 949             channels->rx_count || channels->tx_count || channels->other_count)
 950                 return -EINVAL;
 951
 952         if (!nvdev || nvdev->destroy)
 953                 return -ENODEV;
 954
 955         if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
 956                 return -EINVAL;
 957
 958         if (count > nvdev->max_chn)
 959                 return -EINVAL;
 960
 961         orig = nvdev->num_chn;
 962
 963         memset(&device_info, 0, sizeof(device_info));
 964         device_info.num_chn = count;
 965         device_info.send_sections = nvdev->send_section_cnt;
 966         device_info.send_section_size = nvdev->send_section_size;
 967         device_info.recv_sections = nvdev->recv_section_cnt;
 968         device_info.recv_section_size = nvdev->recv_section_size;
 969
 970         ret = netvsc_detach(net, nvdev);
 971         if (ret)
 972                 return ret;
 973
 974         ret = netvsc_attach(net, &device_info);
 975         if (ret) {
 976                 device_info.num_chn = orig;
 977                 if (netvsc_attach(net, &device_info))
 978                         netdev_err(net, "restoring channel setting failed\n");
 979         }
 980
 981         return ret;
 982 }
 983
 984 static bool
 985 netvsc_validate_ethtool_ss_cmd(const struct ethtool_link_ksettings *cmd)
 986 {
 987         struct ethtool_link_ksettings diff1 = *cmd;
 988         struct ethtool_link_ksettings diff2 = {};
 989
 990         diff1.base.speed = 0;
 991         diff1.base.duplex = 0;
 992         /* advertising and cmd are usually set */
 993         ethtool_link_ksettings_zero_link_mode(&diff1, advertising);
 994         diff1.base.cmd = 0;
 995         /* We set port to PORT_OTHER */
 996         diff2.base.port = PORT_OTHER;
 997
 998         return !memcmp(&diff1, &diff2, sizeof(diff1));
 999 }
1000
1001 static void netvsc_init_settings(struct net_device *dev)
1002 {
1003         struct net_device_context *ndc = netdev_priv(dev);
1004
1005         ndc->l4_hash = HV_DEFAULT_L4HASH;
1006
1007         ndc->speed = SPEED_UNKNOWN;
1008         ndc->duplex = DUPLEX_FULL;
1009 }
1010
1011 static int netvsc_get_link_ksettings(struct net_device *dev,
1012                                      struct ethtool_link_ksettings *cmd)
1013 {
1014         struct net_device_context *ndc = netdev_priv(dev);
1015
1016         cmd->base.speed = ndc->speed;
1017         cmd->base.duplex = ndc->duplex;
1018         cmd->base.port = PORT_OTHER;
1019
1020         return 0;
1021 }
1022
1023 static int netvsc_set_link_ksettings(struct net_device *dev,
1024                                      const struct ethtool_link_ksettings *cmd)
1025 {
1026         struct net_device_context *ndc = netdev_priv(dev);
1027         u32 speed;
1028
1029         speed = cmd->base.speed;
1030         if (!ethtool_validate_speed(speed) ||
1031             !ethtool_validate_duplex(cmd->base.duplex) ||
1032             !netvsc_validate_ethtool_ss_cmd(cmd))
1033                 return -EINVAL;
1034
1035         ndc->speed = speed;
1036         ndc->duplex = cmd->base.duplex;
1037
1038         return 0;
1039 }
1040
1041 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
1042 {
1043         struct net_device_context *ndevctx = netdev_priv(ndev);
1044         struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1045         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1046         int orig_mtu = ndev->mtu;
1047         struct netvsc_device_info device_info;
1048         int ret = 0;
1049
1050         if (!nvdev || nvdev->destroy)
1051                 return -ENODEV;
1052
1053         /* Change MTU of underlying VF netdev first. */
1054         if (vf_netdev) {
1055                 ret = dev_set_mtu(vf_netdev, mtu);
1056                 if (ret)
1057                         return ret;
1058         }
1059
1060         memset(&device_info, 0, sizeof(device_info));
1061         device_info.num_chn = nvdev->num_chn;
1062         device_info.send_sections = nvdev->send_section_cnt;
1063         device_info.send_section_size = nvdev->send_section_size;
1064         device_info.recv_sections = nvdev->recv_section_cnt;
1065         device_info.recv_section_size = nvdev->recv_section_size;
1066
1067         ret = netvsc_detach(ndev, nvdev);
1068         if (ret)
1069                 goto rollback_vf;
1070
1071         ndev->mtu = mtu;
1072
1073         ret = netvsc_attach(ndev, &device_info);
1074         if (ret)
1075                 goto rollback;
1076
1077         return 0;
1078
1079 rollback:
1080         /* Attempt rollback to original MTU */
1081         ndev->mtu = orig_mtu;
1082
1083         if (netvsc_attach(ndev, &device_info))
1084                 netdev_err(ndev, "restoring mtu failed\n");
1085 rollback_vf:
1086         if (vf_netdev)
1087                 dev_set_mtu(vf_netdev, orig_mtu);
1088
1089         return ret;
1090 }
1091
1092 static void netvsc_get_vf_stats(struct net_device *net,
1093                                 struct netvsc_vf_pcpu_stats *tot)
1094 {
1095         struct net_device_context *ndev_ctx = netdev_priv(net);
1096         int i;
1097
1098         memset(tot, 0, sizeof(*tot));
1099
1100         for_each_possible_cpu(i) {
1101                 const struct netvsc_vf_pcpu_stats *stats
1102                         = per_cpu_ptr(ndev_ctx->vf_stats, i);
1103                 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1104                 unsigned int start;
1105
1106                 do {
1107                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1108                         rx_packets = stats->rx_packets;
1109                         tx_packets = stats->tx_packets;
1110                         rx_bytes = stats->rx_bytes;
1111                         tx_bytes = stats->tx_bytes;
1112                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1113
1114                 tot->rx_packets += rx_packets;
1115                 tot->tx_packets += tx_packets;
1116                 tot->rx_bytes   += rx_bytes;
1117                 tot->tx_bytes   += tx_bytes;
1118                 tot->tx_dropped += stats->tx_dropped;
1119         }
1120 }
1121
1122 static void netvsc_get_pcpu_stats(struct net_device *net,
1123                                   struct netvsc_ethtool_pcpu_stats *pcpu_tot)
1124 {
1125         struct net_device_context *ndev_ctx = netdev_priv(net);
1126         struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
1127         int i;
1128
1129         /* fetch percpu stats of vf */
1130         for_each_possible_cpu(i) {
1131                 const struct netvsc_vf_pcpu_stats *stats =
1132                         per_cpu_ptr(ndev_ctx->vf_stats, i);
1133                 struct netvsc_ethtool_pcpu_stats *this_tot = &pcpu_tot[i];
1134                 unsigned int start;
1135
1136                 do {
1137                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1138                         this_tot->vf_rx_packets = stats->rx_packets;
1139                         this_tot->vf_tx_packets = stats->tx_packets;
1140                         this_tot->vf_rx_bytes = stats->rx_bytes;
1141                         this_tot->vf_tx_bytes = stats->tx_bytes;
1142                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1143                 this_tot->rx_packets = this_tot->vf_rx_packets;
1144                 this_tot->tx_packets = this_tot->vf_tx_packets;
1145                 this_tot->rx_bytes   = this_tot->vf_rx_bytes;
1146                 this_tot->tx_bytes   = this_tot->vf_tx_bytes;
1147         }
1148
1149         /* fetch percpu stats of netvsc */
1150         for (i = 0; i < nvdev->num_chn; i++) {
1151                 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1152                 const struct netvsc_stats *stats;
1153                 struct netvsc_ethtool_pcpu_stats *this_tot =
1154                         &pcpu_tot[nvchan->channel->target_cpu];
1155                 u64 packets, bytes;
1156                 unsigned int start;
1157
1158                 stats = &nvchan->tx_stats;
1159                 do {
1160                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1161                         packets = stats->packets;
1162                         bytes = stats->bytes;
1163                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1164
1165                 this_tot->tx_bytes      += bytes;
1166                 this_tot->tx_packets    += packets;
1167
1168                 stats = &nvchan->rx_stats;
1169                 do {
1170                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1171                         packets = stats->packets;
1172                         bytes = stats->bytes;
1173                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1174
1175                 this_tot->rx_bytes      += bytes;
1176                 this_tot->rx_packets    += packets;
1177         }
1178 }
1179
1180 static void netvsc_get_stats64(struct net_device *net,
1181                                struct rtnl_link_stats64 *t)
1182 {
1183         struct net_device_context *ndev_ctx = netdev_priv(net);
1184         struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
1185         struct netvsc_vf_pcpu_stats vf_tot;
1186         int i;
1187
1188         if (!nvdev)
1189                 return;
1190
1191         netdev_stats_to_stats64(t, &net->stats);
1192
1193         netvsc_get_vf_stats(net, &vf_tot);
1194         t->rx_packets += vf_tot.rx_packets;
1195         t->tx_packets += vf_tot.tx_packets;
1196         t->rx_bytes   += vf_tot.rx_bytes;
1197         t->tx_bytes   += vf_tot.tx_bytes;
1198         t->tx_dropped += vf_tot.tx_dropped;
1199
1200         for (i = 0; i < nvdev->num_chn; i++) {
1201                 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1202                 const struct netvsc_stats *stats;
1203                 u64 packets, bytes, multicast;
1204                 unsigned int start;
1205
1206                 stats = &nvchan->tx_stats;
1207                 do {
1208                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1209                         packets = stats->packets;
1210                         bytes = stats->bytes;
1211                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1212
1213                 t->tx_bytes     += bytes;
1214                 t->tx_packets   += packets;
1215
1216                 stats = &nvchan->rx_stats;
1217                 do {
1218                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1219                         packets = stats->packets;
1220                         bytes = stats->bytes;
1221                         multicast = stats->multicast + stats->broadcast;
1222                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1223
1224                 t->rx_bytes     += bytes;
1225                 t->rx_packets   += packets;
1226                 t->multicast    += multicast;
1227         }
1228 }
1229
1230 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
1231 {
1232         struct net_device_context *ndc = netdev_priv(ndev);
1233         struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1234         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1235         struct sockaddr *addr = p;
1236         int err;
1237
1238         err = eth_prepare_mac_addr_change(ndev, p);
1239         if (err)
1240                 return err;
1241
1242         if (!nvdev)
1243                 return -ENODEV;
1244
1245         if (vf_netdev) {
1246                 err = dev_set_mac_address(vf_netdev, addr);
1247                 if (err)
1248                         return err;
1249         }
1250
1251         err = rndis_filter_set_device_mac(nvdev, addr->sa_data);
1252         if (!err) {
1253                 eth_commit_mac_addr_change(ndev, p);
1254         } else if (vf_netdev) {
1255                 /* rollback change on VF */
1256                 memcpy(addr->sa_data, ndev->dev_addr, ETH_ALEN);
1257                 dev_set_mac_address(vf_netdev, addr);
1258         }
1259
1260         return err;
1261 }
1262
1263 static const struct {
1264         char name[ETH_GSTRING_LEN];
1265         u16 offset;
1266 } netvsc_stats[] = {
1267         { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
1268         { "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
1269         { "tx_no_space",  offsetof(struct netvsc_ethtool_stats, tx_no_space) },
1270         { "tx_too_big",   offsetof(struct netvsc_ethtool_stats, tx_too_big) },
1271         { "tx_busy",      offsetof(struct netvsc_ethtool_stats, tx_busy) },
1272         { "tx_send_full", offsetof(struct netvsc_ethtool_stats, tx_send_full) },
1273         { "rx_comp_busy", offsetof(struct netvsc_ethtool_stats, rx_comp_busy) },
1274         { "rx_no_memory", offsetof(struct netvsc_ethtool_stats, rx_no_memory) },
1275         { "stop_queue", offsetof(struct netvsc_ethtool_stats, stop_queue) },
1276         { "wake_queue", offsetof(struct netvsc_ethtool_stats, wake_queue) },
1277 }, pcpu_stats[] = {
1278         { "cpu%u_rx_packets",
1279                 offsetof(struct netvsc_ethtool_pcpu_stats, rx_packets) },
1280         { "cpu%u_rx_bytes",
1281                 offsetof(struct netvsc_ethtool_pcpu_stats, rx_bytes) },
1282         { "cpu%u_tx_packets",
1283                 offsetof(struct netvsc_ethtool_pcpu_stats, tx_packets) },
1284         { "cpu%u_tx_bytes",
1285                 offsetof(struct netvsc_ethtool_pcpu_stats, tx_bytes) },
1286         { "cpu%u_vf_rx_packets",
1287                 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_packets) },
1288         { "cpu%u_vf_rx_bytes",
1289                 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_bytes) },
1290         { "cpu%u_vf_tx_packets",
1291                 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_packets) },
1292         { "cpu%u_vf_tx_bytes",
1293                 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_bytes) },
1294 }, vf_stats[] = {
1295         { "vf_rx_packets", offsetof(struct netvsc_vf_pcpu_stats, rx_packets) },
1296         { "vf_rx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, rx_bytes) },
1297         { "vf_tx_packets", offsetof(struct netvsc_vf_pcpu_stats, tx_packets) },
1298         { "vf_tx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, tx_bytes) },
1299         { "vf_tx_dropped", offsetof(struct netvsc_vf_pcpu_stats, tx_dropped) },
1300 };
1301
1302 #define NETVSC_GLOBAL_STATS_LEN ARRAY_SIZE(netvsc_stats)
1303 #define NETVSC_VF_STATS_LEN     ARRAY_SIZE(vf_stats)
1304
1305 /* statistics per queue (rx/tx packets/bytes) */
1306 #define NETVSC_PCPU_STATS_LEN (num_present_cpus() * ARRAY_SIZE(pcpu_stats))
1307
1308 /* 4 statistics per queue (rx/tx packets/bytes) */
1309 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 4)
1310
1311 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
1312 {
1313         struct net_device_context *ndc = netdev_priv(dev);
1314         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1315
1316         if (!nvdev)
1317                 return -ENODEV;
1318
1319         switch (string_set) {
1320         case ETH_SS_STATS:
1321                 return NETVSC_GLOBAL_STATS_LEN
1322                         + NETVSC_VF_STATS_LEN
1323                         + NETVSC_QUEUE_STATS_LEN(nvdev)
1324                         + NETVSC_PCPU_STATS_LEN;
1325         default:
1326                 return -EINVAL;
1327         }
1328 }
1329
1330 static void netvsc_get_ethtool_stats(struct net_device *dev,
1331                                      struct ethtool_stats *stats, u64 *data)
1332 {
1333         struct net_device_context *ndc = netdev_priv(dev);
1334         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1335         const void *nds = &ndc->eth_stats;
1336         const struct netvsc_stats *qstats;
1337         struct netvsc_vf_pcpu_stats sum;
1338         struct netvsc_ethtool_pcpu_stats *pcpu_sum;
1339         unsigned int start;
1340         u64 packets, bytes;
1341         int i, j, cpu;
1342
1343         if (!nvdev)
1344                 return;
1345
1346         for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
1347                 data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1348
1349         netvsc_get_vf_stats(dev, &sum);
1350         for (j = 0; j < NETVSC_VF_STATS_LEN; j++)
1351                 data[i++] = *(u64 *)((void *)&sum + vf_stats[j].offset);
1352
1353         for (j = 0; j < nvdev->num_chn; j++) {
1354                 qstats = &nvdev->chan_table[j].tx_stats;
1355
1356                 do {
1357                         start = u64_stats_fetch_begin_irq(&qstats->syncp);
1358                         packets = qstats->packets;
1359                         bytes = qstats->bytes;
1360                 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1361                 data[i++] = packets;
1362                 data[i++] = bytes;
1363
1364                 qstats = &nvdev->chan_table[j].rx_stats;
1365                 do {
1366                         start = u64_stats_fetch_begin_irq(&qstats->syncp);
1367                         packets = qstats->packets;
1368                         bytes = qstats->bytes;
1369                 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1370                 data[i++] = packets;
1371                 data[i++] = bytes;
1372         }
1373
1374         pcpu_sum = kvmalloc_array(num_possible_cpus(),
1375                                   sizeof(struct netvsc_ethtool_pcpu_stats),
1376                                   GFP_KERNEL);
1377         netvsc_get_pcpu_stats(dev, pcpu_sum);
1378         for_each_present_cpu(cpu) {
1379                 struct netvsc_ethtool_pcpu_stats *this_sum = &pcpu_sum[cpu];
1380
1381                 for (j = 0; j < ARRAY_SIZE(pcpu_stats); j++)
1382                         data[i++] = *(u64 *)((void *)this_sum
1383                                              + pcpu_stats[j].offset);
1384         }
1385         kvfree(pcpu_sum);
1386 }
1387
1388 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1389 {
1390         struct net_device_context *ndc = netdev_priv(dev);
1391         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1392         u8 *p = data;
1393         int i, cpu;
1394
1395         if (!nvdev)
1396                 return;
1397
1398         switch (stringset) {
1399         case ETH_SS_STATS:
1400                 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++) {
1401                         memcpy(p, netvsc_stats[i].name, ETH_GSTRING_LEN);
1402                         p += ETH_GSTRING_LEN;
1403                 }
1404
1405                 for (i = 0; i < ARRAY_SIZE(vf_stats); i++) {
1406                         memcpy(p, vf_stats[i].name, ETH_GSTRING_LEN);
1407                         p += ETH_GSTRING_LEN;
1408                 }
1409
1410                 for (i = 0; i < nvdev->num_chn; i++) {
1411                         sprintf(p, "tx_queue_%u_packets", i);
1412                         p += ETH_GSTRING_LEN;
1413                         sprintf(p, "tx_queue_%u_bytes", i);
1414                         p += ETH_GSTRING_LEN;
1415                         sprintf(p, "rx_queue_%u_packets", i);
1416                         p += ETH_GSTRING_LEN;
1417                         sprintf(p, "rx_queue_%u_bytes", i);
1418                         p += ETH_GSTRING_LEN;
1419                 }
1420
1421                 for_each_present_cpu(cpu) {
1422                         for (i = 0; i < ARRAY_SIZE(pcpu_stats); i++) {
1423                                 sprintf(p, pcpu_stats[i].name, cpu);
1424                                 p += ETH_GSTRING_LEN;
1425                         }
1426                 }
1427
1428                 break;
1429         }
1430 }
1431
1432 static int
1433 netvsc_get_rss_hash_opts(struct net_device_context *ndc,
1434                          struct ethtool_rxnfc *info)
1435 {
1436         const u32 l4_flag = RXH_L4_B_0_1 | RXH_L4_B_2_3;
1437
1438         info->data = RXH_IP_SRC | RXH_IP_DST;
1439
1440         switch (info->flow_type) {
1441         case TCP_V4_FLOW:
1442                 if (ndc->l4_hash & HV_TCP4_L4HASH)
1443                         info->data |= l4_flag;
1444
1445                 break;
1446
1447         case TCP_V6_FLOW:
1448                 if (ndc->l4_hash & HV_TCP6_L4HASH)
1449                         info->data |= l4_flag;
1450
1451                 break;
1452
1453         case UDP_V4_FLOW:
1454                 if (ndc->l4_hash & HV_UDP4_L4HASH)
1455                         info->data |= l4_flag;
1456
1457                 break;
1458
1459         case UDP_V6_FLOW:
1460                 if (ndc->l4_hash & HV_UDP6_L4HASH)
1461                         info->data |= l4_flag;
1462
1463                 break;
1464
1465         case IPV4_FLOW:
1466         case IPV6_FLOW:
1467                 break;
1468         default:
1469                 info->data = 0;
1470                 break;
1471         }
1472
1473         return 0;
1474 }
1475
1476 static int
1477 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1478                  u32 *rules)
1479 {
1480         struct net_device_context *ndc = netdev_priv(dev);
1481         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1482
1483         if (!nvdev)
1484                 return -ENODEV;
1485
1486         switch (info->cmd) {
1487         case ETHTOOL_GRXRINGS:
1488                 info->data = nvdev->num_chn;
1489                 return 0;
1490
1491         case ETHTOOL_GRXFH:
1492                 return netvsc_get_rss_hash_opts(ndc, info);
1493         }
1494         return -EOPNOTSUPP;
1495 }
1496
1497 static int netvsc_set_rss_hash_opts(struct net_device_context *ndc,
1498                                     struct ethtool_rxnfc *info)
1499 {
1500         if (info->data == (RXH_IP_SRC | RXH_IP_DST |
1501                            RXH_L4_B_0_1 | RXH_L4_B_2_3)) {
1502                 switch (info->flow_type) {
1503                 case TCP_V4_FLOW:
1504                         ndc->l4_hash |= HV_TCP4_L4HASH;
1505                         break;
1506
1507                 case TCP_V6_FLOW:
1508                         ndc->l4_hash |= HV_TCP6_L4HASH;
1509                         break;
1510
1511                 case UDP_V4_FLOW:
1512                         ndc->l4_hash |= HV_UDP4_L4HASH;
1513                         break;
1514
1515                 case UDP_V6_FLOW:
1516                         ndc->l4_hash |= HV_UDP6_L4HASH;
1517                         break;
1518
1519                 default:
1520                         return -EOPNOTSUPP;
1521                 }
1522
1523                 return 0;
1524         }
1525
1526         if (info->data == (RXH_IP_SRC | RXH_IP_DST)) {
1527                 switch (info->flow_type) {
1528                 case TCP_V4_FLOW:
1529                         ndc->l4_hash &= ~HV_TCP4_L4HASH;
1530                         break;
1531
1532                 case TCP_V6_FLOW:
1533                         ndc->l4_hash &= ~HV_TCP6_L4HASH;
1534                         break;
1535
1536                 case UDP_V4_FLOW:
1537                         ndc->l4_hash &= ~HV_UDP4_L4HASH;
1538                         break;
1539
1540                 case UDP_V6_FLOW:
1541                         ndc->l4_hash &= ~HV_UDP6_L4HASH;
1542                         break;
1543
1544                 default:
1545                         return -EOPNOTSUPP;
1546                 }
1547
1548                 return 0;
1549         }
1550
1551         return -EOPNOTSUPP;
1552 }
1553
1554 static int
1555 netvsc_set_rxnfc(struct net_device *ndev, struct ethtool_rxnfc *info)
1556 {
1557         struct net_device_context *ndc = netdev_priv(ndev);
1558
1559         if (info->cmd == ETHTOOL_SRXFH)
1560                 return netvsc_set_rss_hash_opts(ndc, info);
1561
1562         return -EOPNOTSUPP;
1563 }
1564
1565 #ifdef CONFIG_NET_POLL_CONTROLLER
1566 static void netvsc_poll_controller(struct net_device *dev)
1567 {
1568         struct net_device_context *ndc = netdev_priv(dev);
1569         struct netvsc_device *ndev;
1570         int i;
1571
1572         rcu_read_lock();
1573         ndev = rcu_dereference(ndc->nvdev);
1574         if (ndev) {
1575                 for (i = 0; i < ndev->num_chn; i++) {
1576                         struct netvsc_channel *nvchan = &ndev->chan_table[i];
1577
1578                         napi_schedule(&nvchan->napi);
1579                 }
1580         }
1581         rcu_read_unlock();
1582 }
1583 #endif
1584
1585 static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1586 {
1587         return NETVSC_HASH_KEYLEN;
1588 }
1589
1590 static u32 netvsc_rss_indir_size(struct net_device *dev)
1591 {
1592         return ITAB_NUM;
1593 }
1594
1595 static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key,
1596                            u8 *hfunc)
1597 {
1598         struct net_device_context *ndc = netdev_priv(dev);
1599         struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1600         struct rndis_device *rndis_dev;
1601         int i;
1602
1603         if (!ndev)
1604                 return -ENODEV;
1605
1606         if (hfunc)
1607                 *hfunc = ETH_RSS_HASH_TOP;      /* Toeplitz */
1608
1609         rndis_dev = ndev->extension;
1610         if (indir) {
1611                 for (i = 0; i < ITAB_NUM; i++)
1612                         indir[i] = rndis_dev->rx_table[i];
1613         }
1614
1615         if (key)
1616                 memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1617
1618         return 0;
1619 }
1620
1621 static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
1622                            const u8 *key, const u8 hfunc)
1623 {
1624         struct net_device_context *ndc = netdev_priv(dev);
1625         struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1626         struct rndis_device *rndis_dev;
1627         int i;
1628
1629         if (!ndev)
1630                 return -ENODEV;
1631
1632         if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
1633                 return -EOPNOTSUPP;
1634
1635         rndis_dev = ndev->extension;
1636         if (indir) {
1637                 for (i = 0; i < ITAB_NUM; i++)
1638                         if (indir[i] >= ndev->num_chn)
1639                                 return -EINVAL;
1640
1641                 for (i = 0; i < ITAB_NUM; i++)
1642                         rndis_dev->rx_table[i] = indir[i];
1643         }
1644
1645         if (!key) {
1646                 if (!indir)
1647                         return 0;
1648
1649                 key = rndis_dev->rss_key;
1650         }
1651
1652         return rndis_filter_set_rss_param(rndis_dev, key);
1653 }
1654
1655 /* Hyper-V RNDIS protocol does not have ring in the HW sense.
1656  * It does have pre-allocated receive area which is divided into sections.
1657  */
1658 static void __netvsc_get_ringparam(struct netvsc_device *nvdev,
1659                                    struct ethtool_ringparam *ring)
1660 {
1661         u32 max_buf_size;
1662
1663         ring->rx_pending = nvdev->recv_section_cnt;
1664         ring->tx_pending = nvdev->send_section_cnt;
1665
1666         if (nvdev->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
1667                 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
1668         else
1669                 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
1670
1671         ring->rx_max_pending = max_buf_size / nvdev->recv_section_size;
1672         ring->tx_max_pending = NETVSC_SEND_BUFFER_SIZE
1673                 / nvdev->send_section_size;
1674 }
1675
1676 static void netvsc_get_ringparam(struct net_device *ndev,
1677                                  struct ethtool_ringparam *ring)
1678 {
1679         struct net_device_context *ndevctx = netdev_priv(ndev);
1680         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1681
1682         if (!nvdev)
1683                 return;
1684
1685         __netvsc_get_ringparam(nvdev, ring);
1686 }
1687
1688 static int netvsc_set_ringparam(struct net_device *ndev,
1689                                 struct ethtool_ringparam *ring)
1690 {
1691         struct net_device_context *ndevctx = netdev_priv(ndev);
1692         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1693         struct netvsc_device_info device_info;
1694         struct ethtool_ringparam orig;
1695         u32 new_tx, new_rx;
1696         int ret = 0;
1697
1698         if (!nvdev || nvdev->destroy)
1699                 return -ENODEV;
1700
1701         memset(&orig, 0, sizeof(orig));
1702         __netvsc_get_ringparam(nvdev, &orig);
1703
1704         new_tx = clamp_t(u32, ring->tx_pending,
1705                          NETVSC_MIN_TX_SECTIONS, orig.tx_max_pending);
1706         new_rx = clamp_t(u32, ring->rx_pending,
1707                          NETVSC_MIN_RX_SECTIONS, orig.rx_max_pending);
1708
1709         if (new_tx == orig.tx_pending &&
1710             new_rx == orig.rx_pending)
1711                 return 0;        /* no change */
1712
1713         memset(&device_info, 0, sizeof(device_info));
1714         device_info.num_chn = nvdev->num_chn;
1715         device_info.send_sections = new_tx;
1716         device_info.send_section_size = nvdev->send_section_size;
1717         device_info.recv_sections = new_rx;
1718         device_info.recv_section_size = nvdev->recv_section_size;
1719
1720         ret = netvsc_detach(ndev, nvdev);
1721         if (ret)
1722                 return ret;
1723
1724         ret = netvsc_attach(ndev, &device_info);
1725         if (ret) {
1726                 device_info.send_sections = orig.tx_pending;
1727                 device_info.recv_sections = orig.rx_pending;
1728
1729                 if (netvsc_attach(ndev, &device_info))
1730                         netdev_err(ndev, "restoring ringparam failed");
1731         }
1732
1733         return ret;
1734 }
1735
1736 static u32 netvsc_get_msglevel(struct net_device *ndev)
1737 {
1738         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1739
1740         return ndev_ctx->msg_enable;
1741 }
1742
1743 static void netvsc_set_msglevel(struct net_device *ndev, u32 val)
1744 {
1745         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1746
1747         ndev_ctx->msg_enable = val;
1748 }
1749
1750 static const struct ethtool_ops ethtool_ops = {
1751         .get_drvinfo    = netvsc_get_drvinfo,
1752         .get_msglevel   = netvsc_get_msglevel,
1753         .set_msglevel   = netvsc_set_msglevel,
1754         .get_link       = ethtool_op_get_link,
1755         .get_ethtool_stats = netvsc_get_ethtool_stats,
1756         .get_sset_count = netvsc_get_sset_count,
1757         .get_strings    = netvsc_get_strings,
1758         .get_channels   = netvsc_get_channels,
1759         .set_channels   = netvsc_set_channels,
1760         .get_ts_info    = ethtool_op_get_ts_info,
1761         .get_rxnfc      = netvsc_get_rxnfc,
1762         .set_rxnfc      = netvsc_set_rxnfc,
1763         .get_rxfh_key_size = netvsc_get_rxfh_key_size,
1764         .get_rxfh_indir_size = netvsc_rss_indir_size,
1765         .get_rxfh       = netvsc_get_rxfh,
1766         .set_rxfh       = netvsc_set_rxfh,
1767         .get_link_ksettings = netvsc_get_link_ksettings,
1768         .set_link_ksettings = netvsc_set_link_ksettings,
1769         .get_ringparam  = netvsc_get_ringparam,
1770         .set_ringparam  = netvsc_set_ringparam,
1771 };
1772
1773 static const struct net_device_ops device_ops = {
1774         .ndo_open =                     netvsc_open,
1775         .ndo_stop =                     netvsc_close,
1776         .ndo_start_xmit =               netvsc_start_xmit,
1777         .ndo_change_rx_flags =          netvsc_change_rx_flags,
1778         .ndo_set_rx_mode =              netvsc_set_rx_mode,
1779         .ndo_change_mtu =               netvsc_change_mtu,
1780         .ndo_validate_addr =            eth_validate_addr,
1781         .ndo_set_mac_address =          netvsc_set_mac_addr,
1782         .ndo_select_queue =             netvsc_select_queue,
1783         .ndo_get_stats64 =              netvsc_get_stats64,
1784 #ifdef CONFIG_NET_POLL_CONTROLLER
1785         .ndo_poll_controller =          netvsc_poll_controller,
1786 #endif
1787 };
1788
1789 /*
1790  * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
1791  * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
1792  * present send GARP packet to network peers with netif_notify_peers().
1793  */
1794 static void netvsc_link_change(struct work_struct *w)
1795 {
1796         struct net_device_context *ndev_ctx =
1797                 container_of(w, struct net_device_context, dwork.work);
1798         struct hv_device *device_obj = ndev_ctx->device_ctx;
1799         struct net_device *net = hv_get_drvdata(device_obj);
1800         struct netvsc_device *net_device;
1801         struct rndis_device *rdev;
1802         struct netvsc_reconfig *event = NULL;
1803         bool notify = false, reschedule = false;
1804         unsigned long flags, next_reconfig, delay;
1805
1806         /* if changes are happening, comeback later */
1807         if (!rtnl_trylock()) {
1808                 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1809                 return;
1810         }
1811
1812         net_device = rtnl_dereference(ndev_ctx->nvdev);
1813         if (!net_device)
1814                 goto out_unlock;
1815
1816         rdev = net_device->extension;
1817
1818         next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
1819         if (time_is_after_jiffies(next_reconfig)) {
1820                 /* link_watch only sends one notification with current state
1821                  * per second, avoid doing reconfig more frequently. Handle
1822                  * wrap around.
1823                  */
1824                 delay = next_reconfig - jiffies;
1825                 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
1826                 schedule_delayed_work(&ndev_ctx->dwork, delay);
1827                 goto out_unlock;
1828         }
1829         ndev_ctx->last_reconfig = jiffies;
1830
1831         spin_lock_irqsave(&ndev_ctx->lock, flags);
1832         if (!list_empty(&ndev_ctx->reconfig_events)) {
1833                 event = list_first_entry(&ndev_ctx->reconfig_events,
1834                                          struct netvsc_reconfig, list);
1835                 list_del(&event->list);
1836                 reschedule = !list_empty(&ndev_ctx->reconfig_events);
1837         }
1838         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1839
1840         if (!event)
1841                 goto out_unlock;
1842
1843         switch (event->event) {
1844                 /* Only the following events are possible due to the check in
1845                  * netvsc_linkstatus_callback()
1846                  */
1847         case RNDIS_STATUS_MEDIA_CONNECT:
1848                 if (rdev->link_state) {
1849                         rdev->link_state = false;
1850                         netif_carrier_on(net);
1851                         netif_tx_wake_all_queues(net);
1852                 } else {
1853                         notify = true;
1854                 }
1855                 kfree(event);
1856                 break;
1857         case RNDIS_STATUS_MEDIA_DISCONNECT:
1858                 if (!rdev->link_state) {
1859                         rdev->link_state = true;
1860                         netif_carrier_off(net);
1861                         netif_tx_stop_all_queues(net);
1862                 }
1863                 kfree(event);
1864                 break;
1865         case RNDIS_STATUS_NETWORK_CHANGE:
1866                 /* Only makes sense if carrier is present */
1867                 if (!rdev->link_state) {
1868                         rdev->link_state = true;
1869                         netif_carrier_off(net);
1870                         netif_tx_stop_all_queues(net);
1871                         event->event = RNDIS_STATUS_MEDIA_CONNECT;
1872                         spin_lock_irqsave(&ndev_ctx->lock, flags);
1873                         list_add(&event->list, &ndev_ctx->reconfig_events);
1874                         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1875                         reschedule = true;
1876                 }
1877                 break;
1878         }
1879
1880         rtnl_unlock();
1881
1882         if (notify)
1883                 netdev_notify_peers(net);
1884
1885         /* link_watch only sends one notification with current state per
1886          * second, handle next reconfig event in 2 seconds.
1887          */
1888         if (reschedule)
1889                 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1890
1891         return;
1892
1893 out_unlock:
1894         rtnl_unlock();
1895 }
1896
1897 static struct net_device *get_netvsc_bymac(const u8 *mac)
1898 {
1899         struct net_device_context *ndev_ctx;
1900
1901         list_for_each_entry(ndev_ctx, &netvsc_dev_list, list) {
1902                 struct net_device *dev = hv_get_drvdata(ndev_ctx->device_ctx);
1903
1904                 if (ether_addr_equal(mac, dev->perm_addr))
1905                         return dev;
1906         }
1907
1908         return NULL;
1909 }
1910
1911 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
1912 {
1913         struct net_device_context *net_device_ctx;
1914         struct net_device *dev;
1915
1916         dev = netdev_master_upper_dev_get(vf_netdev);
1917         if (!dev || dev->netdev_ops != &device_ops)
1918                 return NULL;    /* not a netvsc device */
1919
1920         net_device_ctx = netdev_priv(dev);
1921         if (!rtnl_dereference(net_device_ctx->nvdev))
1922                 return NULL;    /* device is removed */
1923
1924         return dev;
1925 }
1926
1927 /* Called when VF is injecting data into network stack.
1928  * Change the associated network device from VF to netvsc.
1929  * note: already called with rcu_read_lock
1930  */
1931 static rx_handler_result_t netvsc_vf_handle_frame(struct sk_buff **pskb)
1932 {
1933         struct sk_buff *skb = *pskb;
1934         struct net_device *ndev = rcu_dereference(skb->dev->rx_handler_data);
1935         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1936         struct netvsc_vf_pcpu_stats *pcpu_stats
1937                  = this_cpu_ptr(ndev_ctx->vf_stats);
1938
1939         skb->dev = ndev;
1940
1941         u64_stats_update_begin(&pcpu_stats->syncp);
1942         pcpu_stats->rx_packets++;
1943         pcpu_stats->rx_bytes += skb->len;
1944         u64_stats_update_end(&pcpu_stats->syncp);
1945
1946         return RX_HANDLER_ANOTHER;
1947 }
1948
1949 static int netvsc_vf_join(struct net_device *vf_netdev,
1950                           struct net_device *ndev)
1951 {
1952         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1953         int ret;
1954
1955         ret = netdev_rx_handler_register(vf_netdev,
1956                                          netvsc_vf_handle_frame, ndev);
1957         if (ret != 0) {
1958                 netdev_err(vf_netdev,
1959                            "can not register netvsc VF receive handler (err = %d)\n",
1960                            ret);
1961                 goto rx_handler_failed;
1962         }
1963
1964         ret = netdev_master_upper_dev_link(vf_netdev, ndev,
1965                                            NULL, NULL, NULL);
1966         if (ret != 0) {
1967                 netdev_err(vf_netdev,
1968                            "can not set master device %s (err = %d)\n",
1969                            ndev->name, ret);
1970                 goto upper_link_failed;
1971         }
1972
1973         /* set slave flag before open to prevent IPv6 addrconf */
1974         vf_netdev->flags |= IFF_SLAVE;
1975
1976         schedule_delayed_work(&ndev_ctx->vf_takeover, VF_TAKEOVER_INT);
1977
1978         call_netdevice_notifiers(NETDEV_JOIN, vf_netdev);
1979
1980         netdev_info(vf_netdev, "joined to %s\n", ndev->name);
1981         return 0;
1982
1983 upper_link_failed:
1984         netdev_rx_handler_unregister(vf_netdev);
1985 rx_handler_failed:
1986         return ret;
1987 }
1988
1989 static void __netvsc_vf_setup(struct net_device *ndev,
1990                               struct net_device *vf_netdev)
1991 {
1992         int ret;
1993
1994         /* Align MTU of VF with master */
1995         ret = dev_set_mtu(vf_netdev, ndev->mtu);
1996         if (ret)
1997                 netdev_warn(vf_netdev,
1998                             "unable to change mtu to %u\n", ndev->mtu);
1999
2000         /* set multicast etc flags on VF */
2001         dev_change_flags(vf_netdev, ndev->flags | IFF_SLAVE);
2002
2003         /* sync address list from ndev to VF */
2004         netif_addr_lock_bh(ndev);
2005         dev_uc_sync(vf_netdev, ndev);
2006         dev_mc_sync(vf_netdev, ndev);
2007         netif_addr_unlock_bh(ndev);
2008
2009         if (netif_running(ndev)) {
2010                 ret = dev_open(vf_netdev);
2011                 if (ret)
2012                         netdev_warn(vf_netdev,
2013                                     "unable to open: %d\n", ret);
2014         }
2015 }
2016
2017 /* Setup VF as slave of the synthetic device.
2018  * Runs in workqueue to avoid recursion in netlink callbacks.
2019  */
2020 static void netvsc_vf_setup(struct work_struct *w)
2021 {
2022         struct net_device_context *ndev_ctx
2023                 = container_of(w, struct net_device_context, vf_takeover.work);
2024         struct net_device *ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2025         struct net_device *vf_netdev;
2026
2027         if (!rtnl_trylock()) {
2028                 schedule_delayed_work(&ndev_ctx->vf_takeover, 0);
2029                 return;
2030         }
2031
2032         vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2033         if (vf_netdev)
2034                 __netvsc_vf_setup(ndev, vf_netdev);
2035
2036         rtnl_unlock();
2037 }
2038
2039 static int netvsc_register_vf(struct net_device *vf_netdev)
2040 {
2041         struct net_device *ndev;
2042         struct net_device_context *net_device_ctx;
2043         struct device *pdev = vf_netdev->dev.parent;
2044         struct netvsc_device *netvsc_dev;
2045         int ret;
2046
2047         if (vf_netdev->addr_len != ETH_ALEN)
2048                 return NOTIFY_DONE;
2049
2050         if (!pdev || !dev_is_pci(pdev) || dev_is_pf(pdev))
2051                 return NOTIFY_DONE;
2052
2053         /*
2054          * We will use the MAC address to locate the synthetic interface to
2055          * associate with the VF interface. If we don't find a matching
2056          * synthetic interface, move on.
2057          */
2058         ndev = get_netvsc_bymac(vf_netdev->perm_addr);
2059         if (!ndev)
2060                 return NOTIFY_DONE;
2061
2062         net_device_ctx = netdev_priv(ndev);
2063         netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2064         if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
2065                 return NOTIFY_DONE;
2066
2067         /* if syntihetic interface is a different namespace,
2068          * then move the VF to that namespace; join will be
2069          * done again in that context.
2070          */
2071         if (!net_eq(dev_net(ndev), dev_net(vf_netdev))) {
2072                 ret = dev_change_net_namespace(vf_netdev,
2073                                                dev_net(ndev), "eth%d");
2074                 if (ret)
2075                         netdev_err(vf_netdev,
2076                                    "could not move to same namespace as %s: %d\n",
2077                                    ndev->name, ret);
2078                 else
2079                         netdev_info(vf_netdev,
2080                                     "VF moved to namespace with: %s\n",
2081                                     ndev->name);
2082                 return NOTIFY_DONE;
2083         }
2084
2085         netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
2086
2087         if (netvsc_vf_join(vf_netdev, ndev) != 0)
2088                 return NOTIFY_DONE;
2089
2090         dev_hold(vf_netdev);
2091         rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
2092         return NOTIFY_OK;
2093 }
2094
2095 /* VF up/down change detected, schedule to change data path */
2096 static int netvsc_vf_changed(struct net_device *vf_netdev)
2097 {
2098         struct net_device_context *net_device_ctx;
2099         struct netvsc_device *netvsc_dev;
2100         struct net_device *ndev;
2101         bool vf_is_up = netif_running(vf_netdev);
2102
2103         ndev = get_netvsc_byref(vf_netdev);
2104         if (!ndev)
2105                 return NOTIFY_DONE;
2106
2107         net_device_ctx = netdev_priv(ndev);
2108         netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2109         if (!netvsc_dev)
2110                 return NOTIFY_DONE;
2111
2112         netvsc_switch_datapath(ndev, vf_is_up);
2113         netdev_info(ndev, "Data path switched %s VF: %s\n",
2114                     vf_is_up ? "to" : "from", vf_netdev->name);
2115
2116         return NOTIFY_OK;
2117 }
2118
2119 static int netvsc_unregister_vf(struct net_device *vf_netdev)
2120 {
2121         struct net_device *ndev;
2122         struct net_device_context *net_device_ctx;
2123
2124         ndev = get_netvsc_byref(vf_netdev);
2125         if (!ndev)
2126                 return NOTIFY_DONE;
2127
2128         net_device_ctx = netdev_priv(ndev);
2129         cancel_delayed_work_sync(&net_device_ctx->vf_takeover);
2130
2131         netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
2132
2133         netdev_rx_handler_unregister(vf_netdev);
2134         netdev_upper_dev_unlink(vf_netdev, ndev);
2135         RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
2136         dev_put(vf_netdev);
2137
2138         return NOTIFY_OK;
2139 }
2140
2141 static int netvsc_probe(struct hv_device *dev,
2142                         const struct hv_vmbus_device_id *dev_id)
2143 {
2144         struct net_device *net = NULL;
2145         struct net_device_context *net_device_ctx;
2146         struct netvsc_device_info device_info;
2147         struct netvsc_device *nvdev;
2148         int ret = -ENOMEM;
2149
2150         net = alloc_etherdev_mq(sizeof(struct net_device_context),
2151                                 VRSS_CHANNEL_MAX);
2152         if (!net)
2153                 goto no_net;
2154
2155         netif_carrier_off(net);
2156
2157         netvsc_init_settings(net);
2158
2159         net_device_ctx = netdev_priv(net);
2160         net_device_ctx->device_ctx = dev;
2161         net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
2162         if (netif_msg_probe(net_device_ctx))
2163                 netdev_dbg(net, "netvsc msg_enable: %d\n",
2164                            net_device_ctx->msg_enable);
2165
2166         hv_set_drvdata(dev, net);
2167
2168         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
2169
2170         spin_lock_init(&net_device_ctx->lock);
2171         INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
2172         INIT_DELAYED_WORK(&net_device_ctx->vf_takeover, netvsc_vf_setup);
2173
2174         net_device_ctx->vf_stats
2175                 = netdev_alloc_pcpu_stats(struct netvsc_vf_pcpu_stats);
2176         if (!net_device_ctx->vf_stats)
2177                 goto no_stats;
2178
2179         net->netdev_ops = &device_ops;
2180         net->ethtool_ops = &ethtool_ops;
2181         SET_NETDEV_DEV(net, &dev->device);
2182
2183         /* We always need headroom for rndis header */
2184         net->needed_headroom = RNDIS_AND_PPI_SIZE;
2185
2186         /* Initialize the number of queues to be 1, we may change it if more
2187          * channels are offered later.
2188          */
2189         netif_set_real_num_tx_queues(net, 1);
2190         netif_set_real_num_rx_queues(net, 1);
2191
2192         /* Notify the netvsc driver of the new device */
2193         memset(&device_info, 0, sizeof(device_info));
2194         device_info.num_chn = VRSS_CHANNEL_DEFAULT;
2195         device_info.send_sections = NETVSC_DEFAULT_TX;
2196         device_info.send_section_size = NETVSC_SEND_SECTION_SIZE;
2197         device_info.recv_sections = NETVSC_DEFAULT_RX;
2198         device_info.recv_section_size = NETVSC_RECV_SECTION_SIZE;
2199
2200         nvdev = rndis_filter_device_add(dev, &device_info);
2201         if (IS_ERR(nvdev)) {
2202                 ret = PTR_ERR(nvdev);
2203                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
2204                 goto rndis_failed;
2205         }
2206
2207         memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
2208
2209         /* We must get rtnl lock before scheduling nvdev->subchan_work,
2210          * otherwise netvsc_subchan_work() can get rtnl lock first and wait
2211          * all subchannels to show up, but that may not happen because
2212          * netvsc_probe() can't get rtnl lock and as a result vmbus_onoffer()
2213          * -> ... -> device_add() -> ... -> __device_attach() can't get
2214          * the device lock, so all the subchannels can't be processed --
2215          * finally netvsc_subchan_work() hangs for ever.
2216          */
2217         rtnl_lock();
2218
2219         if (nvdev->num_chn > 1)
2220                 schedule_work(&nvdev->subchan_work);
2221
2222         /* hw_features computed in rndis_netdev_set_hwcaps() */
2223         net->features = net->hw_features |
2224                 NETIF_F_HIGHDMA | NETIF_F_SG |
2225                 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
2226         net->vlan_features = net->features;
2227
2228         netdev_lockdep_set_classes(net);
2229
2230         /* MTU range: 68 - 1500 or 65521 */
2231         net->min_mtu = NETVSC_MTU_MIN;
2232         if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
2233                 net->max_mtu = NETVSC_MTU - ETH_HLEN;
2234         else
2235                 net->max_mtu = ETH_DATA_LEN;
2236
2237         ret = register_netdevice(net);
2238         if (ret != 0) {
2239                 pr_err("Unable to register netdev.\n");
2240                 goto register_failed;
2241         }
2242
2243         list_add(&net_device_ctx->list, &netvsc_dev_list);
2244         rtnl_unlock();
2245         return 0;
2246
2247 register_failed:
2248         rtnl_unlock();
2249         rndis_filter_device_remove(dev, nvdev);
2250 rndis_failed:
2251         free_percpu(net_device_ctx->vf_stats);
2252 no_stats:
2253         hv_set_drvdata(dev, NULL);
2254         free_netdev(net);
2255 no_net:
2256         return ret;
2257 }
2258
2259 static int netvsc_remove(struct hv_device *dev)
2260 {
2261         struct net_device_context *ndev_ctx;
2262         struct net_device *vf_netdev, *net;
2263         struct netvsc_device *nvdev;
2264
2265         net = hv_get_drvdata(dev);
2266         if (net == NULL) {
2267                 dev_err(&dev->device, "No net device to remove\n");
2268                 return 0;
2269         }
2270
2271         ndev_ctx = netdev_priv(net);
2272
2273         cancel_delayed_work_sync(&ndev_ctx->dwork);
2274
2275         rcu_read_lock();
2276         nvdev = rcu_dereference(ndev_ctx->nvdev);
2277
2278         if  (nvdev)
2279                 cancel_work_sync(&nvdev->subchan_work);
2280
2281         /*
2282          * Call to the vsc driver to let it know that the device is being
2283          * removed. Also blocks mtu and channel changes.
2284          */
2285         rtnl_lock();
2286         vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2287         if (vf_netdev)
2288                 netvsc_unregister_vf(vf_netdev);
2289
2290         if (nvdev)
2291                 rndis_filter_device_remove(dev, nvdev);
2292
2293         unregister_netdevice(net);
2294         list_del(&ndev_ctx->list);
2295
2296         rtnl_unlock();
2297         rcu_read_unlock();
2298
2299         hv_set_drvdata(dev, NULL);
2300
2301         free_percpu(ndev_ctx->vf_stats);
2302         free_netdev(net);
2303         return 0;
2304 }
2305
2306 static const struct hv_vmbus_device_id id_table[] = {
2307         /* Network guid */
2308         { HV_NIC_GUID, },
2309         { },
2310 };
2311
2312 MODULE_DEVICE_TABLE(vmbus, id_table);
2313
2314 /* The one and only one */
2315 static struct  hv_driver netvsc_drv = {
2316         .name = KBUILD_MODNAME,
2317         .id_table = id_table,
2318         .probe = netvsc_probe,
2319         .remove = netvsc_remove,
2320         .driver = {
2321                 .probe_type = PROBE_PREFER_ASYNCHRONOUS,
2322         },
2323 };
2324
2325 /*
2326  * On Hyper-V, every VF interface is matched with a corresponding
2327  * synthetic interface. The synthetic interface is presented first
2328  * to the guest. When the corresponding VF instance is registered,
2329  * we will take care of switching the data path.
2330  */
2331 static int netvsc_netdev_event(struct notifier_block *this,
2332                                unsigned long event, void *ptr)
2333 {
2334         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2335
2336         /* Skip our own events */
2337         if (event_dev->netdev_ops == &device_ops)
2338                 return NOTIFY_DONE;
2339
2340         /* Avoid non-Ethernet type devices */
2341         if (event_dev->type != ARPHRD_ETHER)
2342                 return NOTIFY_DONE;
2343
2344         /* Avoid Vlan dev with same MAC registering as VF */
2345         if (is_vlan_dev(event_dev))
2346                 return NOTIFY_DONE;
2347
2348         /* Avoid Bonding master dev with same MAC registering as VF */
2349         if ((event_dev->priv_flags & IFF_BONDING) &&
2350             (event_dev->flags & IFF_MASTER))
2351                 return NOTIFY_DONE;
2352
2353         switch (event) {
2354         case NETDEV_REGISTER:
2355                 return netvsc_register_vf(event_dev);
2356         case NETDEV_UNREGISTER:
2357                 return netvsc_unregister_vf(event_dev);
2358         case NETDEV_UP:
2359         case NETDEV_DOWN:
2360                 return netvsc_vf_changed(event_dev);
2361         default:
2362                 return NOTIFY_DONE;
2363         }
2364 }
2365
2366 static struct notifier_block netvsc_netdev_notifier = {
2367         .notifier_call = netvsc_netdev_event,
2368 };
2369
2370 static void __exit netvsc_drv_exit(void)
2371 {
2372         unregister_netdevice_notifier(&netvsc_netdev_notifier);
2373         vmbus_driver_unregister(&netvsc_drv);
2374 }
2375
2376 static int __init netvsc_drv_init(void)
2377 {
2378         int ret;
2379
2380         if (ring_size < RING_SIZE_MIN) {
2381                 ring_size = RING_SIZE_MIN;
2382                 pr_info("Increased ring_size to %u (min allowed)\n",
2383                         ring_size);
2384         }
2385         netvsc_ring_bytes = ring_size * PAGE_SIZE;
2386
2387         ret = vmbus_driver_register(&netvsc_drv);
2388         if (ret)
2389                 return ret;
2390
2391         register_netdevice_notifier(&netvsc_netdev_notifier);
2392         return 0;
2393 }
2394
2395 MODULE_LICENSE("GPL");
2396 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
2397
2398 module_init(netvsc_drv_init);
2399 module_exit(netvsc_drv_exit);