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, NULL);
 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->internal = 0;
 230         ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
 231
 232         rndis_pkt->per_pkt_info_len += ppi_size;
 233
 234         return ppi + 1;
 235 }
 236
 237 /* Azure hosts don't support non-TCP port numbers in hashing for fragmented
 238  * packets. We can use ethtool to change UDP hash level when necessary.
 239  */
 240 static inline u32 netvsc_get_hash(
 241         struct sk_buff *skb,
 242         const struct net_device_context *ndc)
 243 {
 244         struct flow_keys flow;
 245         u32 hash, pkt_proto = 0;
 246         static u32 hashrnd __read_mostly;
 247
 248         net_get_random_once(&hashrnd, sizeof(hashrnd));
 249
 250         if (!skb_flow_dissect_flow_keys(skb, &flow, 0))
 251                 return 0;
 252
 253         switch (flow.basic.ip_proto) {
 254         case IPPROTO_TCP:
 255                 if (flow.basic.n_proto == htons(ETH_P_IP))
 256                         pkt_proto = HV_TCP4_L4HASH;
 257                 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
 258                         pkt_proto = HV_TCP6_L4HASH;
 259
 260                 break;
 261
 262         case IPPROTO_UDP:
 263                 if (flow.basic.n_proto == htons(ETH_P_IP))
 264                         pkt_proto = HV_UDP4_L4HASH;
 265                 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
 266                         pkt_proto = HV_UDP6_L4HASH;
 267
 268                 break;
 269         }
 270
 271         if (pkt_proto & ndc->l4_hash) {
 272                 return skb_get_hash(skb);
 273         } else {
 274                 if (flow.basic.n_proto == htons(ETH_P_IP))
 275                         hash = jhash2((u32 *)&flow.addrs.v4addrs, 2, hashrnd);
 276                 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
 277                         hash = jhash2((u32 *)&flow.addrs.v6addrs, 8, hashrnd);
 278                 else
 279                         hash = 0;
 280
 281                 skb_set_hash(skb, hash, PKT_HASH_TYPE_L3);
 282         }
 283
 284         return hash;
 285 }
 286
 287 static inline int netvsc_get_tx_queue(struct net_device *ndev,
 288                                       struct sk_buff *skb, int old_idx)
 289 {
 290         const struct net_device_context *ndc = netdev_priv(ndev);
 291         struct sock *sk = skb->sk;
 292         int q_idx;
 293
 294         q_idx = ndc->tx_table[netvsc_get_hash(skb, ndc) &
 295                               (VRSS_SEND_TAB_SIZE - 1)];
 296
 297         /* If queue index changed record the new value */
 298         if (q_idx != old_idx &&
 299             sk && sk_fullsock(sk) && rcu_access_pointer(sk->sk_dst_cache))
 300                 sk_tx_queue_set(sk, q_idx);
 301
 302         return q_idx;
 303 }
 304
 305 /*
 306  * Select queue for transmit.
 307  *
 308  * If a valid queue has already been assigned, then use that.
 309  * Otherwise compute tx queue based on hash and the send table.
 310  *
 311  * This is basically similar to default (netdev_pick_tx) with the added step
 312  * of using the host send_table when no other queue has been assigned.
 313  *
 314  * TODO support XPS - but get_xps_queue not exported
 315  */
 316 static u16 netvsc_pick_tx(struct net_device *ndev, struct sk_buff *skb)
 317 {
 318         int q_idx = sk_tx_queue_get(skb->sk);
 319
 320         if (q_idx < 0 || skb->ooo_okay || q_idx >= ndev->real_num_tx_queues) {
 321                 /* If forwarding a packet, we use the recorded queue when
 322                  * available for better cache locality.
 323                  */
 324                 if (skb_rx_queue_recorded(skb))
 325                         q_idx = skb_get_rx_queue(skb);
 326                 else
 327                         q_idx = netvsc_get_tx_queue(ndev, skb, q_idx);
 328         }
 329
 330         return q_idx;
 331 }
 332
 333 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
 334                                struct net_device *sb_dev)
 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, sb_dev);
 347                 else
 348                         txq = netdev_pick_tx(vf_netdev, skb, NULL);
 349
 350                 /* Record the queue selected by VF so that it can be
 351                  * used for common case where VF has more queues than
 352                  * the synthetic device.
 353                  */
 354                 qdisc_skb_cb(skb)->slave_dev_queue_mapping = txq;
 355         } else {
 356                 txq = netvsc_pick_tx(ndev, skb);
 357         }
 358         rcu_read_unlock();
 359
 360         while (unlikely(txq >= ndev->real_num_tx_queues))
 361                 txq -= ndev->real_num_tx_queues;
 362
 363         return txq;
 364 }
 365
 366 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
 367                        struct hv_page_buffer *pb)
 368 {
 369         int j = 0;
 370
 371         /* Deal with compound pages by ignoring unused part
 372          * of the page.
 373          */
 374         page += (offset >> PAGE_SHIFT);
 375         offset &= ~PAGE_MASK;
 376
 377         while (len > 0) {
 378                 unsigned long bytes;
 379
 380                 bytes = PAGE_SIZE - offset;
 381                 if (bytes > len)
 382                         bytes = len;
 383                 pb[j].pfn = page_to_pfn(page);
 384                 pb[j].offset = offset;
 385                 pb[j].len = bytes;
 386
 387                 offset += bytes;
 388                 len -= bytes;
 389
 390                 if (offset == PAGE_SIZE && len) {
 391                         page++;
 392                         offset = 0;
 393                         j++;
 394                 }
 395         }
 396
 397         return j + 1;
 398 }
 399
 400 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
 401                            struct hv_netvsc_packet *packet,
 402                            struct hv_page_buffer *pb)
 403 {
 404         u32 slots_used = 0;
 405         char *data = skb->data;
 406         int frags = skb_shinfo(skb)->nr_frags;
 407         int i;
 408
 409         /* The packet is laid out thus:
 410          * 1. hdr: RNDIS header and PPI
 411          * 2. skb linear data
 412          * 3. skb fragment data
 413          */
 414         slots_used += fill_pg_buf(virt_to_page(hdr),
 415                                   offset_in_page(hdr),
 416                                   len, &pb[slots_used]);
 417
 418         packet->rmsg_size = len;
 419         packet->rmsg_pgcnt = slots_used;
 420
 421         slots_used += fill_pg_buf(virt_to_page(data),
 422                                 offset_in_page(data),
 423                                 skb_headlen(skb), &pb[slots_used]);
 424
 425         for (i = 0; i < frags; i++) {
 426                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 427
 428                 slots_used += fill_pg_buf(skb_frag_page(frag),
 429                                         frag->page_offset,
 430                                         skb_frag_size(frag), &pb[slots_used]);
 431         }
 432         return slots_used;
 433 }
 434
 435 static int count_skb_frag_slots(struct sk_buff *skb)
 436 {
 437         int i, frags = skb_shinfo(skb)->nr_frags;
 438         int pages = 0;
 439
 440         for (i = 0; i < frags; i++) {
 441                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 442                 unsigned long size = skb_frag_size(frag);
 443                 unsigned long offset = frag->page_offset;
 444
 445                 /* Skip unused frames from start of page */
 446                 offset &= ~PAGE_MASK;
 447                 pages += PFN_UP(offset + size);
 448         }
 449         return pages;
 450 }
 451
 452 static int netvsc_get_slots(struct sk_buff *skb)
 453 {
 454         char *data = skb->data;
 455         unsigned int offset = offset_in_page(data);
 456         unsigned int len = skb_headlen(skb);
 457         int slots;
 458         int frag_slots;
 459
 460         slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
 461         frag_slots = count_skb_frag_slots(skb);
 462         return slots + frag_slots;
 463 }
 464
 465 static u32 net_checksum_info(struct sk_buff *skb)
 466 {
 467         if (skb->protocol == htons(ETH_P_IP)) {
 468                 struct iphdr *ip = ip_hdr(skb);
 469
 470                 if (ip->protocol == IPPROTO_TCP)
 471                         return TRANSPORT_INFO_IPV4_TCP;
 472                 else if (ip->protocol == IPPROTO_UDP)
 473                         return TRANSPORT_INFO_IPV4_UDP;
 474         } else {
 475                 struct ipv6hdr *ip6 = ipv6_hdr(skb);
 476
 477                 if (ip6->nexthdr == IPPROTO_TCP)
 478                         return TRANSPORT_INFO_IPV6_TCP;
 479                 else if (ip6->nexthdr == IPPROTO_UDP)
 480                         return TRANSPORT_INFO_IPV6_UDP;
 481         }
 482
 483         return TRANSPORT_INFO_NOT_IP;
 484 }
 485
 486 /* Send skb on the slave VF device. */
 487 static int netvsc_vf_xmit(struct net_device *net, struct net_device *vf_netdev,
 488                           struct sk_buff *skb)
 489 {
 490         struct net_device_context *ndev_ctx = netdev_priv(net);
 491         unsigned int len = skb->len;
 492         int rc;
 493
 494         skb->dev = vf_netdev;
 495         skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
 496
 497         rc = dev_queue_xmit(skb);
 498         if (likely(rc == NET_XMIT_SUCCESS || rc == NET_XMIT_CN)) {
 499                 struct netvsc_vf_pcpu_stats *pcpu_stats
 500                         = this_cpu_ptr(ndev_ctx->vf_stats);
 501
 502                 u64_stats_update_begin(&pcpu_stats->syncp);
 503                 pcpu_stats->tx_packets++;
 504                 pcpu_stats->tx_bytes += len;
 505                 u64_stats_update_end(&pcpu_stats->syncp);
 506         } else {
 507                 this_cpu_inc(ndev_ctx->vf_stats->tx_dropped);
 508         }
 509
 510         return rc;
 511 }
 512
 513 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
 514 {
 515         struct net_device_context *net_device_ctx = netdev_priv(net);
 516         struct hv_netvsc_packet *packet = NULL;
 517         int ret;
 518         unsigned int num_data_pgs;
 519         struct rndis_message *rndis_msg;
 520         struct net_device *vf_netdev;
 521         u32 rndis_msg_size;
 522         u32 hash;
 523         struct hv_page_buffer pb[MAX_PAGE_BUFFER_COUNT];
 524
 525         /* if VF is present and up then redirect packets
 526          * already called with rcu_read_lock_bh
 527          */
 528         vf_netdev = rcu_dereference_bh(net_device_ctx->vf_netdev);
 529         if (vf_netdev && netif_running(vf_netdev) &&
 530             !netpoll_tx_running(net))
 531                 return netvsc_vf_xmit(net, vf_netdev, skb);
 532
 533         /* We will atmost need two pages to describe the rndis
 534          * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
 535          * of pages in a single packet. If skb is scattered around
 536          * more pages we try linearizing it.
 537          */
 538
 539         num_data_pgs = netvsc_get_slots(skb) + 2;
 540
 541         if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
 542                 ++net_device_ctx->eth_stats.tx_scattered;
 543
 544                 if (skb_linearize(skb))
 545                         goto no_memory;
 546
 547                 num_data_pgs = netvsc_get_slots(skb) + 2;
 548                 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
 549                         ++net_device_ctx->eth_stats.tx_too_big;
 550                         goto drop;
 551                 }
 552         }
 553
 554         /*
 555          * Place the rndis header in the skb head room and
 556          * the skb->cb will be used for hv_netvsc_packet
 557          * structure.
 558          */
 559         ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
 560         if (ret)
 561                 goto no_memory;
 562
 563         /* Use the skb control buffer for building up the packet */
 564         BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
 565                         FIELD_SIZEOF(struct sk_buff, cb));
 566         packet = (struct hv_netvsc_packet *)skb->cb;
 567
 568         packet->q_idx = skb_get_queue_mapping(skb);
 569
 570         packet->total_data_buflen = skb->len;
 571         packet->total_bytes = skb->len;
 572         packet->total_packets = 1;
 573
 574         rndis_msg = (struct rndis_message *)skb->head;
 575
 576         /* Add the rndis header */
 577         rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
 578         rndis_msg->msg_len = packet->total_data_buflen;
 579
 580         rndis_msg->msg.pkt = (struct rndis_packet) {
 581                 .data_offset = sizeof(struct rndis_packet),
 582                 .data_len = packet->total_data_buflen,
 583                 .per_pkt_info_offset = sizeof(struct rndis_packet),
 584         };
 585
 586         rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
 587
 588         hash = skb_get_hash_raw(skb);
 589         if (hash != 0 && net->real_num_tx_queues > 1) {
 590                 u32 *hash_info;
 591
 592                 rndis_msg_size += NDIS_HASH_PPI_SIZE;
 593                 hash_info = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
 594                                           NBL_HASH_VALUE);
 595                 *hash_info = hash;
 596         }
 597
 598         if (skb_vlan_tag_present(skb)) {
 599                 struct ndis_pkt_8021q_info *vlan;
 600
 601                 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
 602                 vlan = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
 603                                      IEEE_8021Q_INFO);
 604
 605                 vlan->value = 0;
 606                 vlan->vlanid = skb_vlan_tag_get_id(skb);
 607                 vlan->cfi = skb_vlan_tag_get_cfi(skb);
 608                 vlan->pri = skb_vlan_tag_get_prio(skb);
 609         }
 610
 611         if (skb_is_gso(skb)) {
 612                 struct ndis_tcp_lso_info *lso_info;
 613
 614                 rndis_msg_size += NDIS_LSO_PPI_SIZE;
 615                 lso_info = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
 616                                          TCP_LARGESEND_PKTINFO);
 617
 618                 lso_info->value = 0;
 619                 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
 620                 if (skb->protocol == htons(ETH_P_IP)) {
 621                         lso_info->lso_v2_transmit.ip_version =
 622                                 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
 623                         ip_hdr(skb)->tot_len = 0;
 624                         ip_hdr(skb)->check = 0;
 625                         tcp_hdr(skb)->check =
 626                                 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
 627                                                    ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 628                 } else {
 629                         lso_info->lso_v2_transmit.ip_version =
 630                                 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
 631                         ipv6_hdr(skb)->payload_len = 0;
 632                         tcp_hdr(skb)->check =
 633                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 634                                                  &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 635                 }
 636                 lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb);
 637                 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
 638         } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
 639                 if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) {
 640                         struct ndis_tcp_ip_checksum_info *csum_info;
 641
 642                         rndis_msg_size += NDIS_CSUM_PPI_SIZE;
 643                         csum_info = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
 644                                                   TCPIP_CHKSUM_PKTINFO);
 645
 646                         csum_info->value = 0;
 647                         csum_info->transmit.tcp_header_offset = skb_transport_offset(skb);
 648
 649                         if (skb->protocol == htons(ETH_P_IP)) {
 650                                 csum_info->transmit.is_ipv4 = 1;
 651
 652                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
 653                                         csum_info->transmit.tcp_checksum = 1;
 654                                 else
 655                                         csum_info->transmit.udp_checksum = 1;
 656                         } else {
 657                                 csum_info->transmit.is_ipv6 = 1;
 658
 659                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
 660                                         csum_info->transmit.tcp_checksum = 1;
 661                                 else
 662                                         csum_info->transmit.udp_checksum = 1;
 663                         }
 664                 } else {
 665                         /* Can't do offload of this type of checksum */
 666                         if (skb_checksum_help(skb))
 667                                 goto drop;
 668                 }
 669         }
 670
 671         /* Start filling in the page buffers with the rndis hdr */
 672         rndis_msg->msg_len += rndis_msg_size;
 673         packet->total_data_buflen = rndis_msg->msg_len;
 674         packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
 675                                                skb, packet, pb);
 676
 677         /* timestamp packet in software */
 678         skb_tx_timestamp(skb);
 679
 680         ret = netvsc_send(net, packet, rndis_msg, pb, skb);
 681         if (likely(ret == 0))
 682                 return NETDEV_TX_OK;
 683
 684         if (ret == -EAGAIN) {
 685                 ++net_device_ctx->eth_stats.tx_busy;
 686                 return NETDEV_TX_BUSY;
 687         }
 688
 689         if (ret == -ENOSPC)
 690                 ++net_device_ctx->eth_stats.tx_no_space;
 691
 692 drop:
 693         dev_kfree_skb_any(skb);
 694         net->stats.tx_dropped++;
 695
 696         return NETDEV_TX_OK;
 697
 698 no_memory:
 699         ++net_device_ctx->eth_stats.tx_no_memory;
 700         goto drop;
 701 }
 702
 703 /*
 704  * netvsc_linkstatus_callback - Link up/down notification
 705  */
 706 void netvsc_linkstatus_callback(struct net_device *net,
 707                                 struct rndis_message *resp)
 708 {
 709         struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
 710         struct net_device_context *ndev_ctx = netdev_priv(net);
 711         struct netvsc_reconfig *event;
 712         unsigned long flags;
 713
 714         /* Update the physical link speed when changing to another vSwitch */
 715         if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
 716                 u32 speed;
 717
 718                 speed = *(u32 *)((void *)indicate
 719                                  + indicate->status_buf_offset) / 10000;
 720                 ndev_ctx->speed = speed;
 721                 return;
 722         }
 723
 724         /* Handle these link change statuses below */
 725         if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
 726             indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
 727             indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
 728                 return;
 729
 730         if (net->reg_state != NETREG_REGISTERED)
 731                 return;
 732
 733         event = kzalloc(sizeof(*event), GFP_ATOMIC);
 734         if (!event)
 735                 return;
 736         event->event = indicate->status;
 737
 738         spin_lock_irqsave(&ndev_ctx->lock, flags);
 739         list_add_tail(&event->list, &ndev_ctx->reconfig_events);
 740         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
 741
 742         schedule_delayed_work(&ndev_ctx->dwork, 0);
 743 }
 744
 745 static void netvsc_comp_ipcsum(struct sk_buff *skb)
 746 {
 747         struct iphdr *iph = (struct iphdr *)skb->data;
 748
 749         iph->check = 0;
 750         iph->check = ip_fast_csum(iph, iph->ihl);
 751 }
 752
 753 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
 754                                              struct netvsc_channel *nvchan)
 755 {
 756         struct napi_struct *napi = &nvchan->napi;
 757         const struct ndis_pkt_8021q_info *vlan = nvchan->rsc.vlan;
 758         const struct ndis_tcp_ip_checksum_info *csum_info =
 759                                                 nvchan->rsc.csum_info;
 760         struct sk_buff *skb;
 761         int i;
 762
 763         skb = napi_alloc_skb(napi, nvchan->rsc.pktlen);
 764         if (!skb)
 765                 return skb;
 766
 767         /*
 768          * Copy to skb. This copy is needed here since the memory pointed by
 769          * hv_netvsc_packet cannot be deallocated
 770          */
 771         for (i = 0; i < nvchan->rsc.cnt; i++)
 772                 skb_put_data(skb, nvchan->rsc.data[i], nvchan->rsc.len[i]);
 773
 774         skb->protocol = eth_type_trans(skb, net);
 775
 776         /* skb is already created with CHECKSUM_NONE */
 777         skb_checksum_none_assert(skb);
 778
 779         /* Incoming packets may have IP header checksum verified by the host.
 780          * They may not have IP header checksum computed after coalescing.
 781          * We compute it here if the flags are set, because on Linux, the IP
 782          * checksum is always checked.
 783          */
 784         if (csum_info && csum_info->receive.ip_checksum_value_invalid &&
 785             csum_info->receive.ip_checksum_succeeded &&
 786             skb->protocol == htons(ETH_P_IP))
 787                 netvsc_comp_ipcsum(skb);
 788
 789         /* Do L4 checksum offload if enabled and present.
 790          */
 791         if (csum_info && (net->features & NETIF_F_RXCSUM)) {
 792                 if (csum_info->receive.tcp_checksum_succeeded ||
 793                     csum_info->receive.udp_checksum_succeeded)
 794                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 795         }
 796
 797         if (vlan) {
 798                 u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT) |
 799                         (vlan->cfi ? VLAN_CFI_MASK : 0);
 800
 801                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 802                                        vlan_tci);
 803         }
 804
 805         return skb;
 806 }
 807
 808 /*
 809  * netvsc_recv_callback -  Callback when we receive a packet from the
 810  * "wire" on the specified device.
 811  */
 812 int netvsc_recv_callback(struct net_device *net,
 813                          struct netvsc_device *net_device,
 814                          struct netvsc_channel *nvchan)
 815 {
 816         struct net_device_context *net_device_ctx = netdev_priv(net);
 817         struct vmbus_channel *channel = nvchan->channel;
 818         u16 q_idx = channel->offermsg.offer.sub_channel_index;
 819         struct sk_buff *skb;
 820         struct netvsc_stats *rx_stats;
 821
 822         if (net->reg_state != NETREG_REGISTERED)
 823                 return NVSP_STAT_FAIL;
 824
 825         /* Allocate a skb - TODO direct I/O to pages? */
 826         skb = netvsc_alloc_recv_skb(net, nvchan);
 827
 828         if (unlikely(!skb)) {
 829                 ++net_device_ctx->eth_stats.rx_no_memory;
 830                 rcu_read_unlock();
 831                 return NVSP_STAT_FAIL;
 832         }
 833
 834         skb_record_rx_queue(skb, q_idx);
 835
 836         /*
 837          * Even if injecting the packet, record the statistics
 838          * on the synthetic device because modifying the VF device
 839          * statistics will not work correctly.
 840          */
 841         rx_stats = &nvchan->rx_stats;
 842         u64_stats_update_begin(&rx_stats->syncp);
 843         rx_stats->packets++;
 844         rx_stats->bytes += nvchan->rsc.pktlen;
 845
 846         if (skb->pkt_type == PACKET_BROADCAST)
 847                 ++rx_stats->broadcast;
 848         else if (skb->pkt_type == PACKET_MULTICAST)
 849                 ++rx_stats->multicast;
 850         u64_stats_update_end(&rx_stats->syncp);
 851
 852         napi_gro_receive(&nvchan->napi, skb);
 853         return NVSP_STAT_SUCCESS;
 854 }
 855
 856 static void netvsc_get_drvinfo(struct net_device *net,
 857                                struct ethtool_drvinfo *info)
 858 {
 859         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
 860         strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
 861 }
 862
 863 static void netvsc_get_channels(struct net_device *net,
 864                                 struct ethtool_channels *channel)
 865 {
 866         struct net_device_context *net_device_ctx = netdev_priv(net);
 867         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
 868
 869         if (nvdev) {
 870                 channel->max_combined   = nvdev->max_chn;
 871                 channel->combined_count = nvdev->num_chn;
 872         }
 873 }
 874
 875 /* Alloc struct netvsc_device_info, and initialize it from either existing
 876  * struct netvsc_device, or from default values.
 877  */
 878 static struct netvsc_device_info *netvsc_devinfo_get
 879                         (struct netvsc_device *nvdev)
 880 {
 881         struct netvsc_device_info *dev_info;
 882
 883         dev_info = kzalloc(sizeof(*dev_info), GFP_ATOMIC);
 884
 885         if (!dev_info)
 886                 return NULL;
 887
 888         if (nvdev) {
 889                 dev_info->num_chn = nvdev->num_chn;
 890                 dev_info->send_sections = nvdev->send_section_cnt;
 891                 dev_info->send_section_size = nvdev->send_section_size;
 892                 dev_info->recv_sections = nvdev->recv_section_cnt;
 893                 dev_info->recv_section_size = nvdev->recv_section_size;
 894
 895                 memcpy(dev_info->rss_key, nvdev->extension->rss_key,
 896                        NETVSC_HASH_KEYLEN);
 897         } else {
 898                 dev_info->num_chn = VRSS_CHANNEL_DEFAULT;
 899                 dev_info->send_sections = NETVSC_DEFAULT_TX;
 900                 dev_info->send_section_size = NETVSC_SEND_SECTION_SIZE;
 901                 dev_info->recv_sections = NETVSC_DEFAULT_RX;
 902                 dev_info->recv_section_size = NETVSC_RECV_SECTION_SIZE;
 903         }
 904
 905         return dev_info;
 906 }
 907
 908 static int netvsc_detach(struct net_device *ndev,
 909                          struct netvsc_device *nvdev)
 910 {
 911         struct net_device_context *ndev_ctx = netdev_priv(ndev);
 912         struct hv_device *hdev = ndev_ctx->device_ctx;
 913         int ret;
 914
 915         /* Don't try continuing to try and setup sub channels */
 916         if (cancel_work_sync(&nvdev->subchan_work))
 917                 nvdev->num_chn = 1;
 918
 919         /* If device was up (receiving) then shutdown */
 920         if (netif_running(ndev)) {
 921                 netif_tx_disable(ndev);
 922
 923                 ret = rndis_filter_close(nvdev);
 924                 if (ret) {
 925                         netdev_err(ndev,
 926                                    "unable to close device (ret %d).\n", ret);
 927                         return ret;
 928                 }
 929
 930                 ret = netvsc_wait_until_empty(nvdev);
 931                 if (ret) {
 932                         netdev_err(ndev,
 933                                    "Ring buffer not empty after closing rndis\n");
 934                         return ret;
 935                 }
 936         }
 937
 938         netif_device_detach(ndev);
 939
 940         rndis_filter_device_remove(hdev, nvdev);
 941
 942         return 0;
 943 }
 944
 945 static int netvsc_attach(struct net_device *ndev,
 946                          struct netvsc_device_info *dev_info)
 947 {
 948         struct net_device_context *ndev_ctx = netdev_priv(ndev);
 949         struct hv_device *hdev = ndev_ctx->device_ctx;
 950         struct netvsc_device *nvdev;
 951         struct rndis_device *rdev;
 952         int ret;
 953
 954         nvdev = rndis_filter_device_add(hdev, dev_info);
 955         if (IS_ERR(nvdev))
 956                 return PTR_ERR(nvdev);
 957
 958         if (nvdev->num_chn > 1) {
 959                 ret = rndis_set_subchannel(ndev, nvdev, dev_info);
 960
 961                 /* if unavailable, just proceed with one queue */
 962                 if (ret) {
 963                         nvdev->max_chn = 1;
 964                         nvdev->num_chn = 1;
 965                 }
 966         }
 967
 968         /* In any case device is now ready */
 969         netif_device_attach(ndev);
 970
 971         /* Note: enable and attach happen when sub-channels setup */
 972         netif_carrier_off(ndev);
 973
 974         if (netif_running(ndev)) {
 975                 ret = rndis_filter_open(nvdev);
 976                 if (ret)
 977                         return ret;
 978
 979                 rdev = nvdev->extension;
 980                 if (!rdev->link_state)
 981                         netif_carrier_on(ndev);
 982         }
 983
 984         return 0;
 985 }
 986
 987 static int netvsc_set_channels(struct net_device *net,
 988                                struct ethtool_channels *channels)
 989 {
 990         struct net_device_context *net_device_ctx = netdev_priv(net);
 991         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
 992         unsigned int orig, count = channels->combined_count;
 993         struct netvsc_device_info *device_info;
 994         int ret;
 995
 996         /* We do not support separate count for rx, tx, or other */
 997         if (count == 0 ||
 998             channels->rx_count || channels->tx_count || channels->other_count)
 999                 return -EINVAL;
1000
1001         if (!nvdev || nvdev->destroy)
1002                 return -ENODEV;
1003
1004         if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
1005                 return -EINVAL;
1006
1007         if (count > nvdev->max_chn)
1008                 return -EINVAL;
1009
1010         orig = nvdev->num_chn;
1011
1012         device_info = netvsc_devinfo_get(nvdev);
1013
1014         if (!device_info)
1015                 return -ENOMEM;
1016
1017         device_info->num_chn = count;
1018
1019         ret = netvsc_detach(net, nvdev);
1020         if (ret)
1021                 goto out;
1022
1023         ret = netvsc_attach(net, device_info);
1024         if (ret) {
1025                 device_info->num_chn = orig;
1026                 if (netvsc_attach(net, device_info))
1027                         netdev_err(net, "restoring channel setting failed\n");
1028         }
1029
1030 out:
1031         kfree(device_info);
1032         return ret;
1033 }
1034
1035 static bool
1036 netvsc_validate_ethtool_ss_cmd(const struct ethtool_link_ksettings *cmd)
1037 {
1038         struct ethtool_link_ksettings diff1 = *cmd;
1039         struct ethtool_link_ksettings diff2 = {};
1040
1041         diff1.base.speed = 0;
1042         diff1.base.duplex = 0;
1043         /* advertising and cmd are usually set */
1044         ethtool_link_ksettings_zero_link_mode(&diff1, advertising);
1045         diff1.base.cmd = 0;
1046         /* We set port to PORT_OTHER */
1047         diff2.base.port = PORT_OTHER;
1048
1049         return !memcmp(&diff1, &diff2, sizeof(diff1));
1050 }
1051
1052 static void netvsc_init_settings(struct net_device *dev)
1053 {
1054         struct net_device_context *ndc = netdev_priv(dev);
1055
1056         ndc->l4_hash = HV_DEFAULT_L4HASH;
1057
1058         ndc->speed = SPEED_UNKNOWN;
1059         ndc->duplex = DUPLEX_FULL;
1060
1061         dev->features = NETIF_F_LRO;
1062 }
1063
1064 static int netvsc_get_link_ksettings(struct net_device *dev,
1065                                      struct ethtool_link_ksettings *cmd)
1066 {
1067         struct net_device_context *ndc = netdev_priv(dev);
1068
1069         cmd->base.speed = ndc->speed;
1070         cmd->base.duplex = ndc->duplex;
1071         cmd->base.port = PORT_OTHER;
1072
1073         return 0;
1074 }
1075
1076 static int netvsc_set_link_ksettings(struct net_device *dev,
1077                                      const struct ethtool_link_ksettings *cmd)
1078 {
1079         struct net_device_context *ndc = netdev_priv(dev);
1080         u32 speed;
1081
1082         speed = cmd->base.speed;
1083         if (!ethtool_validate_speed(speed) ||
1084             !ethtool_validate_duplex(cmd->base.duplex) ||
1085             !netvsc_validate_ethtool_ss_cmd(cmd))
1086                 return -EINVAL;
1087
1088         ndc->speed = speed;
1089         ndc->duplex = cmd->base.duplex;
1090
1091         return 0;
1092 }
1093
1094 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
1095 {
1096         struct net_device_context *ndevctx = netdev_priv(ndev);
1097         struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1098         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1099         int orig_mtu = ndev->mtu;
1100         struct netvsc_device_info *device_info;
1101         int ret = 0;
1102
1103         if (!nvdev || nvdev->destroy)
1104                 return -ENODEV;
1105
1106         device_info = netvsc_devinfo_get(nvdev);
1107
1108         if (!device_info)
1109                 return -ENOMEM;
1110
1111         /* Change MTU of underlying VF netdev first. */
1112         if (vf_netdev) {
1113                 ret = dev_set_mtu(vf_netdev, mtu);
1114                 if (ret)
1115                         goto out;
1116         }
1117
1118         ret = netvsc_detach(ndev, nvdev);
1119         if (ret)
1120                 goto rollback_vf;
1121
1122         ndev->mtu = mtu;
1123
1124         ret = netvsc_attach(ndev, device_info);
1125         if (!ret)
1126                 goto out;
1127
1128         /* Attempt rollback to original MTU */
1129         ndev->mtu = orig_mtu;
1130
1131         if (netvsc_attach(ndev, device_info))
1132                 netdev_err(ndev, "restoring mtu failed\n");
1133 rollback_vf:
1134         if (vf_netdev)
1135                 dev_set_mtu(vf_netdev, orig_mtu);
1136
1137 out:
1138         kfree(device_info);
1139         return ret;
1140 }
1141
1142 static void netvsc_get_vf_stats(struct net_device *net,
1143                                 struct netvsc_vf_pcpu_stats *tot)
1144 {
1145         struct net_device_context *ndev_ctx = netdev_priv(net);
1146         int i;
1147
1148         memset(tot, 0, sizeof(*tot));
1149
1150         for_each_possible_cpu(i) {
1151                 const struct netvsc_vf_pcpu_stats *stats
1152                         = per_cpu_ptr(ndev_ctx->vf_stats, i);
1153                 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1154                 unsigned int start;
1155
1156                 do {
1157                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1158                         rx_packets = stats->rx_packets;
1159                         tx_packets = stats->tx_packets;
1160                         rx_bytes = stats->rx_bytes;
1161                         tx_bytes = stats->tx_bytes;
1162                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1163
1164                 tot->rx_packets += rx_packets;
1165                 tot->tx_packets += tx_packets;
1166                 tot->rx_bytes   += rx_bytes;
1167                 tot->tx_bytes   += tx_bytes;
1168                 tot->tx_dropped += stats->tx_dropped;
1169         }
1170 }
1171
1172 static void netvsc_get_pcpu_stats(struct net_device *net,
1173                                   struct netvsc_ethtool_pcpu_stats *pcpu_tot)
1174 {
1175         struct net_device_context *ndev_ctx = netdev_priv(net);
1176         struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
1177         int i;
1178
1179         /* fetch percpu stats of vf */
1180         for_each_possible_cpu(i) {
1181                 const struct netvsc_vf_pcpu_stats *stats =
1182                         per_cpu_ptr(ndev_ctx->vf_stats, i);
1183                 struct netvsc_ethtool_pcpu_stats *this_tot = &pcpu_tot[i];
1184                 unsigned int start;
1185
1186                 do {
1187                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1188                         this_tot->vf_rx_packets = stats->rx_packets;
1189                         this_tot->vf_tx_packets = stats->tx_packets;
1190                         this_tot->vf_rx_bytes = stats->rx_bytes;
1191                         this_tot->vf_tx_bytes = stats->tx_bytes;
1192                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1193                 this_tot->rx_packets = this_tot->vf_rx_packets;
1194                 this_tot->tx_packets = this_tot->vf_tx_packets;
1195                 this_tot->rx_bytes   = this_tot->vf_rx_bytes;
1196                 this_tot->tx_bytes   = this_tot->vf_tx_bytes;
1197         }
1198
1199         /* fetch percpu stats of netvsc */
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                 struct netvsc_ethtool_pcpu_stats *this_tot =
1204                         &pcpu_tot[nvchan->channel->target_cpu];
1205                 u64 packets, bytes;
1206                 unsigned int start;
1207
1208                 stats = &nvchan->tx_stats;
1209                 do {
1210                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1211                         packets = stats->packets;
1212                         bytes = stats->bytes;
1213                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1214
1215                 this_tot->tx_bytes      += bytes;
1216                 this_tot->tx_packets    += packets;
1217
1218                 stats = &nvchan->rx_stats;
1219                 do {
1220                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1221                         packets = stats->packets;
1222                         bytes = stats->bytes;
1223                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1224
1225                 this_tot->rx_bytes      += bytes;
1226                 this_tot->rx_packets    += packets;
1227         }
1228 }
1229
1230 static void netvsc_get_stats64(struct net_device *net,
1231                                struct rtnl_link_stats64 *t)
1232 {
1233         struct net_device_context *ndev_ctx = netdev_priv(net);
1234         struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
1235         struct netvsc_vf_pcpu_stats vf_tot;
1236         int i;
1237
1238         if (!nvdev)
1239                 return;
1240
1241         netdev_stats_to_stats64(t, &net->stats);
1242
1243         netvsc_get_vf_stats(net, &vf_tot);
1244         t->rx_packets += vf_tot.rx_packets;
1245         t->tx_packets += vf_tot.tx_packets;
1246         t->rx_bytes   += vf_tot.rx_bytes;
1247         t->tx_bytes   += vf_tot.tx_bytes;
1248         t->tx_dropped += vf_tot.tx_dropped;
1249
1250         for (i = 0; i < nvdev->num_chn; i++) {
1251                 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1252                 const struct netvsc_stats *stats;
1253                 u64 packets, bytes, multicast;
1254                 unsigned int start;
1255
1256                 stats = &nvchan->tx_stats;
1257                 do {
1258                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1259                         packets = stats->packets;
1260                         bytes = stats->bytes;
1261                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1262
1263                 t->tx_bytes     += bytes;
1264                 t->tx_packets   += packets;
1265
1266                 stats = &nvchan->rx_stats;
1267                 do {
1268                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1269                         packets = stats->packets;
1270                         bytes = stats->bytes;
1271                         multicast = stats->multicast + stats->broadcast;
1272                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1273
1274                 t->rx_bytes     += bytes;
1275                 t->rx_packets   += packets;
1276                 t->multicast    += multicast;
1277         }
1278 }
1279
1280 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
1281 {
1282         struct net_device_context *ndc = netdev_priv(ndev);
1283         struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1284         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1285         struct sockaddr *addr = p;
1286         int err;
1287
1288         err = eth_prepare_mac_addr_change(ndev, p);
1289         if (err)
1290                 return err;
1291
1292         if (!nvdev)
1293                 return -ENODEV;
1294
1295         if (vf_netdev) {
1296                 err = dev_set_mac_address(vf_netdev, addr, NULL);
1297                 if (err)
1298                         return err;
1299         }
1300
1301         err = rndis_filter_set_device_mac(nvdev, addr->sa_data);
1302         if (!err) {
1303                 eth_commit_mac_addr_change(ndev, p);
1304         } else if (vf_netdev) {
1305                 /* rollback change on VF */
1306                 memcpy(addr->sa_data, ndev->dev_addr, ETH_ALEN);
1307                 dev_set_mac_address(vf_netdev, addr, NULL);
1308         }
1309
1310         return err;
1311 }
1312
1313 static const struct {
1314         char name[ETH_GSTRING_LEN];
1315         u16 offset;
1316 } netvsc_stats[] = {
1317         { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
1318         { "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
1319         { "tx_no_space",  offsetof(struct netvsc_ethtool_stats, tx_no_space) },
1320         { "tx_too_big",   offsetof(struct netvsc_ethtool_stats, tx_too_big) },
1321         { "tx_busy",      offsetof(struct netvsc_ethtool_stats, tx_busy) },
1322         { "tx_send_full", offsetof(struct netvsc_ethtool_stats, tx_send_full) },
1323         { "rx_comp_busy", offsetof(struct netvsc_ethtool_stats, rx_comp_busy) },
1324         { "rx_no_memory", offsetof(struct netvsc_ethtool_stats, rx_no_memory) },
1325         { "stop_queue", offsetof(struct netvsc_ethtool_stats, stop_queue) },
1326         { "wake_queue", offsetof(struct netvsc_ethtool_stats, wake_queue) },
1327 }, pcpu_stats[] = {
1328         { "cpu%u_rx_packets",
1329                 offsetof(struct netvsc_ethtool_pcpu_stats, rx_packets) },
1330         { "cpu%u_rx_bytes",
1331                 offsetof(struct netvsc_ethtool_pcpu_stats, rx_bytes) },
1332         { "cpu%u_tx_packets",
1333                 offsetof(struct netvsc_ethtool_pcpu_stats, tx_packets) },
1334         { "cpu%u_tx_bytes",
1335                 offsetof(struct netvsc_ethtool_pcpu_stats, tx_bytes) },
1336         { "cpu%u_vf_rx_packets",
1337                 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_packets) },
1338         { "cpu%u_vf_rx_bytes",
1339                 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_bytes) },
1340         { "cpu%u_vf_tx_packets",
1341                 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_packets) },
1342         { "cpu%u_vf_tx_bytes",
1343                 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_bytes) },
1344 }, vf_stats[] = {
1345         { "vf_rx_packets", offsetof(struct netvsc_vf_pcpu_stats, rx_packets) },
1346         { "vf_rx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, rx_bytes) },
1347         { "vf_tx_packets", offsetof(struct netvsc_vf_pcpu_stats, tx_packets) },
1348         { "vf_tx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, tx_bytes) },
1349         { "vf_tx_dropped", offsetof(struct netvsc_vf_pcpu_stats, tx_dropped) },
1350 };
1351
1352 #define NETVSC_GLOBAL_STATS_LEN ARRAY_SIZE(netvsc_stats)
1353 #define NETVSC_VF_STATS_LEN     ARRAY_SIZE(vf_stats)
1354
1355 /* statistics per queue (rx/tx packets/bytes) */
1356 #define NETVSC_PCPU_STATS_LEN (num_present_cpus() * ARRAY_SIZE(pcpu_stats))
1357
1358 /* 4 statistics per queue (rx/tx packets/bytes) */
1359 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 4)
1360
1361 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
1362 {
1363         struct net_device_context *ndc = netdev_priv(dev);
1364         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1365
1366         if (!nvdev)
1367                 return -ENODEV;
1368
1369         switch (string_set) {
1370         case ETH_SS_STATS:
1371                 return NETVSC_GLOBAL_STATS_LEN
1372                         + NETVSC_VF_STATS_LEN
1373                         + NETVSC_QUEUE_STATS_LEN(nvdev)
1374                         + NETVSC_PCPU_STATS_LEN;
1375         default:
1376                 return -EINVAL;
1377         }
1378 }
1379
1380 static void netvsc_get_ethtool_stats(struct net_device *dev,
1381                                      struct ethtool_stats *stats, u64 *data)
1382 {
1383         struct net_device_context *ndc = netdev_priv(dev);
1384         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1385         const void *nds = &ndc->eth_stats;
1386         const struct netvsc_stats *qstats;
1387         struct netvsc_vf_pcpu_stats sum;
1388         struct netvsc_ethtool_pcpu_stats *pcpu_sum;
1389         unsigned int start;
1390         u64 packets, bytes;
1391         int i, j, cpu;
1392
1393         if (!nvdev)
1394                 return;
1395
1396         for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
1397                 data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1398
1399         netvsc_get_vf_stats(dev, &sum);
1400         for (j = 0; j < NETVSC_VF_STATS_LEN; j++)
1401                 data[i++] = *(u64 *)((void *)&sum + vf_stats[j].offset);
1402
1403         for (j = 0; j < nvdev->num_chn; j++) {
1404                 qstats = &nvdev->chan_table[j].tx_stats;
1405
1406                 do {
1407                         start = u64_stats_fetch_begin_irq(&qstats->syncp);
1408                         packets = qstats->packets;
1409                         bytes = qstats->bytes;
1410                 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1411                 data[i++] = packets;
1412                 data[i++] = bytes;
1413
1414                 qstats = &nvdev->chan_table[j].rx_stats;
1415                 do {
1416                         start = u64_stats_fetch_begin_irq(&qstats->syncp);
1417                         packets = qstats->packets;
1418                         bytes = qstats->bytes;
1419                 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1420                 data[i++] = packets;
1421                 data[i++] = bytes;
1422         }
1423
1424         pcpu_sum = kvmalloc_array(num_possible_cpus(),
1425                                   sizeof(struct netvsc_ethtool_pcpu_stats),
1426                                   GFP_KERNEL);
1427         netvsc_get_pcpu_stats(dev, pcpu_sum);
1428         for_each_present_cpu(cpu) {
1429                 struct netvsc_ethtool_pcpu_stats *this_sum = &pcpu_sum[cpu];
1430
1431                 for (j = 0; j < ARRAY_SIZE(pcpu_stats); j++)
1432                         data[i++] = *(u64 *)((void *)this_sum
1433                                              + pcpu_stats[j].offset);
1434         }
1435         kvfree(pcpu_sum);
1436 }
1437
1438 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1439 {
1440         struct net_device_context *ndc = netdev_priv(dev);
1441         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1442         u8 *p = data;
1443         int i, cpu;
1444
1445         if (!nvdev)
1446                 return;
1447
1448         switch (stringset) {
1449         case ETH_SS_STATS:
1450                 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++) {
1451                         memcpy(p, netvsc_stats[i].name, ETH_GSTRING_LEN);
1452                         p += ETH_GSTRING_LEN;
1453                 }
1454
1455                 for (i = 0; i < ARRAY_SIZE(vf_stats); i++) {
1456                         memcpy(p, vf_stats[i].name, ETH_GSTRING_LEN);
1457                         p += ETH_GSTRING_LEN;
1458                 }
1459
1460                 for (i = 0; i < nvdev->num_chn; i++) {
1461                         sprintf(p, "tx_queue_%u_packets", i);
1462                         p += ETH_GSTRING_LEN;
1463                         sprintf(p, "tx_queue_%u_bytes", i);
1464                         p += ETH_GSTRING_LEN;
1465                         sprintf(p, "rx_queue_%u_packets", i);
1466                         p += ETH_GSTRING_LEN;
1467                         sprintf(p, "rx_queue_%u_bytes", i);
1468                         p += ETH_GSTRING_LEN;
1469                 }
1470
1471                 for_each_present_cpu(cpu) {
1472                         for (i = 0; i < ARRAY_SIZE(pcpu_stats); i++) {
1473                                 sprintf(p, pcpu_stats[i].name, cpu);
1474                                 p += ETH_GSTRING_LEN;
1475                         }
1476                 }
1477
1478                 break;
1479         }
1480 }
1481
1482 static int
1483 netvsc_get_rss_hash_opts(struct net_device_context *ndc,
1484                          struct ethtool_rxnfc *info)
1485 {
1486         const u32 l4_flag = RXH_L4_B_0_1 | RXH_L4_B_2_3;
1487
1488         info->data = RXH_IP_SRC | RXH_IP_DST;
1489
1490         switch (info->flow_type) {
1491         case TCP_V4_FLOW:
1492                 if (ndc->l4_hash & HV_TCP4_L4HASH)
1493                         info->data |= l4_flag;
1494
1495                 break;
1496
1497         case TCP_V6_FLOW:
1498                 if (ndc->l4_hash & HV_TCP6_L4HASH)
1499                         info->data |= l4_flag;
1500
1501                 break;
1502
1503         case UDP_V4_FLOW:
1504                 if (ndc->l4_hash & HV_UDP4_L4HASH)
1505                         info->data |= l4_flag;
1506
1507                 break;
1508
1509         case UDP_V6_FLOW:
1510                 if (ndc->l4_hash & HV_UDP6_L4HASH)
1511                         info->data |= l4_flag;
1512
1513                 break;
1514
1515         case IPV4_FLOW:
1516         case IPV6_FLOW:
1517                 break;
1518         default:
1519                 info->data = 0;
1520                 break;
1521         }
1522
1523         return 0;
1524 }
1525
1526 static int
1527 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1528                  u32 *rules)
1529 {
1530         struct net_device_context *ndc = netdev_priv(dev);
1531         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1532
1533         if (!nvdev)
1534                 return -ENODEV;
1535
1536         switch (info->cmd) {
1537         case ETHTOOL_GRXRINGS:
1538                 info->data = nvdev->num_chn;
1539                 return 0;
1540
1541         case ETHTOOL_GRXFH:
1542                 return netvsc_get_rss_hash_opts(ndc, info);
1543         }
1544         return -EOPNOTSUPP;
1545 }
1546
1547 static int netvsc_set_rss_hash_opts(struct net_device_context *ndc,
1548                                     struct ethtool_rxnfc *info)
1549 {
1550         if (info->data == (RXH_IP_SRC | RXH_IP_DST |
1551                            RXH_L4_B_0_1 | RXH_L4_B_2_3)) {
1552                 switch (info->flow_type) {
1553                 case TCP_V4_FLOW:
1554                         ndc->l4_hash |= HV_TCP4_L4HASH;
1555                         break;
1556
1557                 case TCP_V6_FLOW:
1558                         ndc->l4_hash |= HV_TCP6_L4HASH;
1559                         break;
1560
1561                 case UDP_V4_FLOW:
1562                         ndc->l4_hash |= HV_UDP4_L4HASH;
1563                         break;
1564
1565                 case UDP_V6_FLOW:
1566                         ndc->l4_hash |= HV_UDP6_L4HASH;
1567                         break;
1568
1569                 default:
1570                         return -EOPNOTSUPP;
1571                 }
1572
1573                 return 0;
1574         }
1575
1576         if (info->data == (RXH_IP_SRC | RXH_IP_DST)) {
1577                 switch (info->flow_type) {
1578                 case TCP_V4_FLOW:
1579                         ndc->l4_hash &= ~HV_TCP4_L4HASH;
1580                         break;
1581
1582                 case TCP_V6_FLOW:
1583                         ndc->l4_hash &= ~HV_TCP6_L4HASH;
1584                         break;
1585
1586                 case UDP_V4_FLOW:
1587                         ndc->l4_hash &= ~HV_UDP4_L4HASH;
1588                         break;
1589
1590                 case UDP_V6_FLOW:
1591                         ndc->l4_hash &= ~HV_UDP6_L4HASH;
1592                         break;
1593
1594                 default:
1595                         return -EOPNOTSUPP;
1596                 }
1597
1598                 return 0;
1599         }
1600
1601         return -EOPNOTSUPP;
1602 }
1603
1604 static int
1605 netvsc_set_rxnfc(struct net_device *ndev, struct ethtool_rxnfc *info)
1606 {
1607         struct net_device_context *ndc = netdev_priv(ndev);
1608
1609         if (info->cmd == ETHTOOL_SRXFH)
1610                 return netvsc_set_rss_hash_opts(ndc, info);
1611
1612         return -EOPNOTSUPP;
1613 }
1614
1615 static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1616 {
1617         return NETVSC_HASH_KEYLEN;
1618 }
1619
1620 static u32 netvsc_rss_indir_size(struct net_device *dev)
1621 {
1622         return ITAB_NUM;
1623 }
1624
1625 static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key,
1626                            u8 *hfunc)
1627 {
1628         struct net_device_context *ndc = netdev_priv(dev);
1629         struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1630         struct rndis_device *rndis_dev;
1631         int i;
1632
1633         if (!ndev)
1634                 return -ENODEV;
1635
1636         if (hfunc)
1637                 *hfunc = ETH_RSS_HASH_TOP;      /* Toeplitz */
1638
1639         rndis_dev = ndev->extension;
1640         if (indir) {
1641                 for (i = 0; i < ITAB_NUM; i++)
1642                         indir[i] = rndis_dev->rx_table[i];
1643         }
1644
1645         if (key)
1646                 memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1647
1648         return 0;
1649 }
1650
1651 static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
1652                            const u8 *key, const u8 hfunc)
1653 {
1654         struct net_device_context *ndc = netdev_priv(dev);
1655         struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1656         struct rndis_device *rndis_dev;
1657         int i;
1658
1659         if (!ndev)
1660                 return -ENODEV;
1661
1662         if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
1663                 return -EOPNOTSUPP;
1664
1665         rndis_dev = ndev->extension;
1666         if (indir) {
1667                 for (i = 0; i < ITAB_NUM; i++)
1668                         if (indir[i] >= ndev->num_chn)
1669                                 return -EINVAL;
1670
1671                 for (i = 0; i < ITAB_NUM; i++)
1672                         rndis_dev->rx_table[i] = indir[i];
1673         }
1674
1675         if (!key) {
1676                 if (!indir)
1677                         return 0;
1678
1679                 key = rndis_dev->rss_key;
1680         }
1681
1682         return rndis_filter_set_rss_param(rndis_dev, key);
1683 }
1684
1685 /* Hyper-V RNDIS protocol does not have ring in the HW sense.
1686  * It does have pre-allocated receive area which is divided into sections.
1687  */
1688 static void __netvsc_get_ringparam(struct netvsc_device *nvdev,
1689                                    struct ethtool_ringparam *ring)
1690 {
1691         u32 max_buf_size;
1692
1693         ring->rx_pending = nvdev->recv_section_cnt;
1694         ring->tx_pending = nvdev->send_section_cnt;
1695
1696         if (nvdev->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
1697                 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
1698         else
1699                 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
1700
1701         ring->rx_max_pending = max_buf_size / nvdev->recv_section_size;
1702         ring->tx_max_pending = NETVSC_SEND_BUFFER_SIZE
1703                 / nvdev->send_section_size;
1704 }
1705
1706 static void netvsc_get_ringparam(struct net_device *ndev,
1707                                  struct ethtool_ringparam *ring)
1708 {
1709         struct net_device_context *ndevctx = netdev_priv(ndev);
1710         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1711
1712         if (!nvdev)
1713                 return;
1714
1715         __netvsc_get_ringparam(nvdev, ring);
1716 }
1717
1718 static int netvsc_set_ringparam(struct net_device *ndev,
1719                                 struct ethtool_ringparam *ring)
1720 {
1721         struct net_device_context *ndevctx = netdev_priv(ndev);
1722         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1723         struct netvsc_device_info *device_info;
1724         struct ethtool_ringparam orig;
1725         u32 new_tx, new_rx;
1726         int ret = 0;
1727
1728         if (!nvdev || nvdev->destroy)
1729                 return -ENODEV;
1730
1731         memset(&orig, 0, sizeof(orig));
1732         __netvsc_get_ringparam(nvdev, &orig);
1733
1734         new_tx = clamp_t(u32, ring->tx_pending,
1735                          NETVSC_MIN_TX_SECTIONS, orig.tx_max_pending);
1736         new_rx = clamp_t(u32, ring->rx_pending,
1737                          NETVSC_MIN_RX_SECTIONS, orig.rx_max_pending);
1738
1739         if (new_tx == orig.tx_pending &&
1740             new_rx == orig.rx_pending)
1741                 return 0;        /* no change */
1742
1743         device_info = netvsc_devinfo_get(nvdev);
1744
1745         if (!device_info)
1746                 return -ENOMEM;
1747
1748         device_info->send_sections = new_tx;
1749         device_info->recv_sections = new_rx;
1750
1751         ret = netvsc_detach(ndev, nvdev);
1752         if (ret)
1753                 goto out;
1754
1755         ret = netvsc_attach(ndev, device_info);
1756         if (ret) {
1757                 device_info->send_sections = orig.tx_pending;
1758                 device_info->recv_sections = orig.rx_pending;
1759
1760                 if (netvsc_attach(ndev, device_info))
1761                         netdev_err(ndev, "restoring ringparam failed");
1762         }
1763
1764 out:
1765         kfree(device_info);
1766         return ret;
1767 }
1768
1769 static int netvsc_set_features(struct net_device *ndev,
1770                                netdev_features_t features)
1771 {
1772         netdev_features_t change = features ^ ndev->features;
1773         struct net_device_context *ndevctx = netdev_priv(ndev);
1774         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1775         struct ndis_offload_params offloads;
1776
1777         if (!nvdev || nvdev->destroy)
1778                 return -ENODEV;
1779
1780         if (!(change & NETIF_F_LRO))
1781                 return 0;
1782
1783         memset(&offloads, 0, sizeof(struct ndis_offload_params));
1784
1785         if (features & NETIF_F_LRO) {
1786                 offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1787                 offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1788         } else {
1789                 offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1790                 offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1791         }
1792
1793         return rndis_filter_set_offload_params(ndev, nvdev, &offloads);
1794 }
1795
1796 static u32 netvsc_get_msglevel(struct net_device *ndev)
1797 {
1798         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1799
1800         return ndev_ctx->msg_enable;
1801 }
1802
1803 static void netvsc_set_msglevel(struct net_device *ndev, u32 val)
1804 {
1805         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1806
1807         ndev_ctx->msg_enable = val;
1808 }
1809
1810 static const struct ethtool_ops ethtool_ops = {
1811         .get_drvinfo    = netvsc_get_drvinfo,
1812         .get_msglevel   = netvsc_get_msglevel,
1813         .set_msglevel   = netvsc_set_msglevel,
1814         .get_link       = ethtool_op_get_link,
1815         .get_ethtool_stats = netvsc_get_ethtool_stats,
1816         .get_sset_count = netvsc_get_sset_count,
1817         .get_strings    = netvsc_get_strings,
1818         .get_channels   = netvsc_get_channels,
1819         .set_channels   = netvsc_set_channels,
1820         .get_ts_info    = ethtool_op_get_ts_info,
1821         .get_rxnfc      = netvsc_get_rxnfc,
1822         .set_rxnfc      = netvsc_set_rxnfc,
1823         .get_rxfh_key_size = netvsc_get_rxfh_key_size,
1824         .get_rxfh_indir_size = netvsc_rss_indir_size,
1825         .get_rxfh       = netvsc_get_rxfh,
1826         .set_rxfh       = netvsc_set_rxfh,
1827         .get_link_ksettings = netvsc_get_link_ksettings,
1828         .set_link_ksettings = netvsc_set_link_ksettings,
1829         .get_ringparam  = netvsc_get_ringparam,
1830         .set_ringparam  = netvsc_set_ringparam,
1831 };
1832
1833 static const struct net_device_ops device_ops = {
1834         .ndo_open =                     netvsc_open,
1835         .ndo_stop =                     netvsc_close,
1836         .ndo_start_xmit =               netvsc_start_xmit,
1837         .ndo_change_rx_flags =          netvsc_change_rx_flags,
1838         .ndo_set_rx_mode =              netvsc_set_rx_mode,
1839         .ndo_set_features =             netvsc_set_features,
1840         .ndo_change_mtu =               netvsc_change_mtu,
1841         .ndo_validate_addr =            eth_validate_addr,
1842         .ndo_set_mac_address =          netvsc_set_mac_addr,
1843         .ndo_select_queue =             netvsc_select_queue,
1844         .ndo_get_stats64 =              netvsc_get_stats64,
1845 };
1846
1847 /*
1848  * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
1849  * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
1850  * present send GARP packet to network peers with netif_notify_peers().
1851  */
1852 static void netvsc_link_change(struct work_struct *w)
1853 {
1854         struct net_device_context *ndev_ctx =
1855                 container_of(w, struct net_device_context, dwork.work);
1856         struct hv_device *device_obj = ndev_ctx->device_ctx;
1857         struct net_device *net = hv_get_drvdata(device_obj);
1858         struct netvsc_device *net_device;
1859         struct rndis_device *rdev;
1860         struct netvsc_reconfig *event = NULL;
1861         bool notify = false, reschedule = false;
1862         unsigned long flags, next_reconfig, delay;
1863
1864         /* if changes are happening, comeback later */
1865         if (!rtnl_trylock()) {
1866                 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1867                 return;
1868         }
1869
1870         net_device = rtnl_dereference(ndev_ctx->nvdev);
1871         if (!net_device)
1872                 goto out_unlock;
1873
1874         rdev = net_device->extension;
1875
1876         next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
1877         if (time_is_after_jiffies(next_reconfig)) {
1878                 /* link_watch only sends one notification with current state
1879                  * per second, avoid doing reconfig more frequently. Handle
1880                  * wrap around.
1881                  */
1882                 delay = next_reconfig - jiffies;
1883                 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
1884                 schedule_delayed_work(&ndev_ctx->dwork, delay);
1885                 goto out_unlock;
1886         }
1887         ndev_ctx->last_reconfig = jiffies;
1888
1889         spin_lock_irqsave(&ndev_ctx->lock, flags);
1890         if (!list_empty(&ndev_ctx->reconfig_events)) {
1891                 event = list_first_entry(&ndev_ctx->reconfig_events,
1892                                          struct netvsc_reconfig, list);
1893                 list_del(&event->list);
1894                 reschedule = !list_empty(&ndev_ctx->reconfig_events);
1895         }
1896         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1897
1898         if (!event)
1899                 goto out_unlock;
1900
1901         switch (event->event) {
1902                 /* Only the following events are possible due to the check in
1903                  * netvsc_linkstatus_callback()
1904                  */
1905         case RNDIS_STATUS_MEDIA_CONNECT:
1906                 if (rdev->link_state) {
1907                         rdev->link_state = false;
1908                         netif_carrier_on(net);
1909                         netif_tx_wake_all_queues(net);
1910                 } else {
1911                         notify = true;
1912                 }
1913                 kfree(event);
1914                 break;
1915         case RNDIS_STATUS_MEDIA_DISCONNECT:
1916                 if (!rdev->link_state) {
1917                         rdev->link_state = true;
1918                         netif_carrier_off(net);
1919                         netif_tx_stop_all_queues(net);
1920                 }
1921                 kfree(event);
1922                 break;
1923         case RNDIS_STATUS_NETWORK_CHANGE:
1924                 /* Only makes sense if carrier is present */
1925                 if (!rdev->link_state) {
1926                         rdev->link_state = true;
1927                         netif_carrier_off(net);
1928                         netif_tx_stop_all_queues(net);
1929                         event->event = RNDIS_STATUS_MEDIA_CONNECT;
1930                         spin_lock_irqsave(&ndev_ctx->lock, flags);
1931                         list_add(&event->list, &ndev_ctx->reconfig_events);
1932                         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1933                         reschedule = true;
1934                 }
1935                 break;
1936         }
1937
1938         rtnl_unlock();
1939
1940         if (notify)
1941                 netdev_notify_peers(net);
1942
1943         /* link_watch only sends one notification with current state per
1944          * second, handle next reconfig event in 2 seconds.
1945          */
1946         if (reschedule)
1947                 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1948
1949         return;
1950
1951 out_unlock:
1952         rtnl_unlock();
1953 }
1954
1955 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
1956 {
1957         struct net_device_context *net_device_ctx;
1958         struct net_device *dev;
1959
1960         dev = netdev_master_upper_dev_get(vf_netdev);
1961         if (!dev || dev->netdev_ops != &device_ops)
1962                 return NULL;    /* not a netvsc device */
1963
1964         net_device_ctx = netdev_priv(dev);
1965         if (!rtnl_dereference(net_device_ctx->nvdev))
1966                 return NULL;    /* device is removed */
1967
1968         return dev;
1969 }
1970
1971 /* Called when VF is injecting data into network stack.
1972  * Change the associated network device from VF to netvsc.
1973  * note: already called with rcu_read_lock
1974  */
1975 static rx_handler_result_t netvsc_vf_handle_frame(struct sk_buff **pskb)
1976 {
1977         struct sk_buff *skb = *pskb;
1978         struct net_device *ndev = rcu_dereference(skb->dev->rx_handler_data);
1979         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1980         struct netvsc_vf_pcpu_stats *pcpu_stats
1981                  = this_cpu_ptr(ndev_ctx->vf_stats);
1982
1983         skb->dev = ndev;
1984
1985         u64_stats_update_begin(&pcpu_stats->syncp);
1986         pcpu_stats->rx_packets++;
1987         pcpu_stats->rx_bytes += skb->len;
1988         u64_stats_update_end(&pcpu_stats->syncp);
1989
1990         return RX_HANDLER_ANOTHER;
1991 }
1992
1993 static int netvsc_vf_join(struct net_device *vf_netdev,
1994                           struct net_device *ndev)
1995 {
1996         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1997         int ret;
1998
1999         ret = netdev_rx_handler_register(vf_netdev,
2000                                          netvsc_vf_handle_frame, ndev);
2001         if (ret != 0) {
2002                 netdev_err(vf_netdev,
2003                            "can not register netvsc VF receive handler (err = %d)\n",
2004                            ret);
2005                 goto rx_handler_failed;
2006         }
2007
2008         ret = netdev_master_upper_dev_link(vf_netdev, ndev,
2009                                            NULL, NULL, NULL);
2010         if (ret != 0) {
2011                 netdev_err(vf_netdev,
2012                            "can not set master device %s (err = %d)\n",
2013                            ndev->name, ret);
2014                 goto upper_link_failed;
2015         }
2016
2017         /* set slave flag before open to prevent IPv6 addrconf */
2018         vf_netdev->flags |= IFF_SLAVE;
2019
2020         schedule_delayed_work(&ndev_ctx->vf_takeover, VF_TAKEOVER_INT);
2021
2022         call_netdevice_notifiers(NETDEV_JOIN, vf_netdev);
2023
2024         netdev_info(vf_netdev, "joined to %s\n", ndev->name);
2025         return 0;
2026
2027 upper_link_failed:
2028         netdev_rx_handler_unregister(vf_netdev);
2029 rx_handler_failed:
2030         return ret;
2031 }
2032
2033 static void __netvsc_vf_setup(struct net_device *ndev,
2034                               struct net_device *vf_netdev)
2035 {
2036         int ret;
2037
2038         /* Align MTU of VF with master */
2039         ret = dev_set_mtu(vf_netdev, ndev->mtu);
2040         if (ret)
2041                 netdev_warn(vf_netdev,
2042                             "unable to change mtu to %u\n", ndev->mtu);
2043
2044         /* set multicast etc flags on VF */
2045         dev_change_flags(vf_netdev, ndev->flags | IFF_SLAVE, NULL);
2046
2047         /* sync address list from ndev to VF */
2048         netif_addr_lock_bh(ndev);
2049         dev_uc_sync(vf_netdev, ndev);
2050         dev_mc_sync(vf_netdev, ndev);
2051         netif_addr_unlock_bh(ndev);
2052
2053         if (netif_running(ndev)) {
2054                 ret = dev_open(vf_netdev, NULL);
2055                 if (ret)
2056                         netdev_warn(vf_netdev,
2057                                     "unable to open: %d\n", ret);
2058         }
2059 }
2060
2061 /* Setup VF as slave of the synthetic device.
2062  * Runs in workqueue to avoid recursion in netlink callbacks.
2063  */
2064 static void netvsc_vf_setup(struct work_struct *w)
2065 {
2066         struct net_device_context *ndev_ctx
2067                 = container_of(w, struct net_device_context, vf_takeover.work);
2068         struct net_device *ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2069         struct net_device *vf_netdev;
2070
2071         if (!rtnl_trylock()) {
2072                 schedule_delayed_work(&ndev_ctx->vf_takeover, 0);
2073                 return;
2074         }
2075
2076         vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2077         if (vf_netdev)
2078                 __netvsc_vf_setup(ndev, vf_netdev);
2079
2080         rtnl_unlock();
2081 }
2082
2083 /* Find netvsc by VF serial number.
2084  * The PCI hyperv controller records the serial number as the slot kobj name.
2085  */
2086 static struct net_device *get_netvsc_byslot(const struct net_device *vf_netdev)
2087 {
2088         struct device *parent = vf_netdev->dev.parent;
2089         struct net_device_context *ndev_ctx;
2090         struct pci_dev *pdev;
2091         u32 serial;
2092
2093         if (!parent || !dev_is_pci(parent))
2094                 return NULL; /* not a PCI device */
2095
2096         pdev = to_pci_dev(parent);
2097         if (!pdev->slot) {
2098                 netdev_notice(vf_netdev, "no PCI slot information\n");
2099                 return NULL;
2100         }
2101
2102         if (kstrtou32(pci_slot_name(pdev->slot), 10, &serial)) {
2103                 netdev_notice(vf_netdev, "Invalid vf serial:%s\n",
2104                               pci_slot_name(pdev->slot));
2105                 return NULL;
2106         }
2107
2108         list_for_each_entry(ndev_ctx, &netvsc_dev_list, list) {
2109                 if (!ndev_ctx->vf_alloc)
2110                         continue;
2111
2112                 if (ndev_ctx->vf_serial == serial)
2113                         return hv_get_drvdata(ndev_ctx->device_ctx);
2114         }
2115
2116         netdev_notice(vf_netdev,
2117                       "no netdev found for vf serial:%u\n", serial);
2118         return NULL;
2119 }
2120
2121 static int netvsc_register_vf(struct net_device *vf_netdev)
2122 {
2123         struct net_device_context *net_device_ctx;
2124         struct netvsc_device *netvsc_dev;
2125         struct net_device *ndev;
2126         int ret;
2127
2128         if (vf_netdev->addr_len != ETH_ALEN)
2129                 return NOTIFY_DONE;
2130
2131         ndev = get_netvsc_byslot(vf_netdev);
2132         if (!ndev)
2133                 return NOTIFY_DONE;
2134
2135         net_device_ctx = netdev_priv(ndev);
2136         netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2137         if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
2138                 return NOTIFY_DONE;
2139
2140         /* if synthetic interface is a different namespace,
2141          * then move the VF to that namespace; join will be
2142          * done again in that context.
2143          */
2144         if (!net_eq(dev_net(ndev), dev_net(vf_netdev))) {
2145                 ret = dev_change_net_namespace(vf_netdev,
2146                                                dev_net(ndev), "eth%d");
2147                 if (ret)
2148                         netdev_err(vf_netdev,
2149                                    "could not move to same namespace as %s: %d\n",
2150                                    ndev->name, ret);
2151                 else
2152                         netdev_info(vf_netdev,
2153                                     "VF moved to namespace with: %s\n",
2154                                     ndev->name);
2155                 return NOTIFY_DONE;
2156         }
2157
2158         netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
2159
2160         if (netvsc_vf_join(vf_netdev, ndev) != 0)
2161                 return NOTIFY_DONE;
2162
2163         dev_hold(vf_netdev);
2164         rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
2165         return NOTIFY_OK;
2166 }
2167
2168 /* VF up/down change detected, schedule to change data path */
2169 static int netvsc_vf_changed(struct net_device *vf_netdev)
2170 {
2171         struct net_device_context *net_device_ctx;
2172         struct netvsc_device *netvsc_dev;
2173         struct net_device *ndev;
2174         bool vf_is_up = netif_running(vf_netdev);
2175
2176         ndev = get_netvsc_byref(vf_netdev);
2177         if (!ndev)
2178                 return NOTIFY_DONE;
2179
2180         net_device_ctx = netdev_priv(ndev);
2181         netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2182         if (!netvsc_dev)
2183                 return NOTIFY_DONE;
2184
2185         netvsc_switch_datapath(ndev, vf_is_up);
2186         netdev_info(ndev, "Data path switched %s VF: %s\n",
2187                     vf_is_up ? "to" : "from", vf_netdev->name);
2188
2189         return NOTIFY_OK;
2190 }
2191
2192 static int netvsc_unregister_vf(struct net_device *vf_netdev)
2193 {
2194         struct net_device *ndev;
2195         struct net_device_context *net_device_ctx;
2196
2197         ndev = get_netvsc_byref(vf_netdev);
2198         if (!ndev)
2199                 return NOTIFY_DONE;
2200
2201         net_device_ctx = netdev_priv(ndev);
2202         cancel_delayed_work_sync(&net_device_ctx->vf_takeover);
2203
2204         netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
2205
2206         netdev_rx_handler_unregister(vf_netdev);
2207         netdev_upper_dev_unlink(vf_netdev, ndev);
2208         RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
2209         dev_put(vf_netdev);
2210
2211         return NOTIFY_OK;
2212 }
2213
2214 static int netvsc_probe(struct hv_device *dev,
2215                         const struct hv_vmbus_device_id *dev_id)
2216 {
2217         struct net_device *net = NULL;
2218         struct net_device_context *net_device_ctx;
2219         struct netvsc_device_info *device_info = NULL;
2220         struct netvsc_device *nvdev;
2221         int ret = -ENOMEM;
2222
2223         net = alloc_etherdev_mq(sizeof(struct net_device_context),
2224                                 VRSS_CHANNEL_MAX);
2225         if (!net)
2226                 goto no_net;
2227
2228         netif_carrier_off(net);
2229
2230         netvsc_init_settings(net);
2231
2232         net_device_ctx = netdev_priv(net);
2233         net_device_ctx->device_ctx = dev;
2234         net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
2235         if (netif_msg_probe(net_device_ctx))
2236                 netdev_dbg(net, "netvsc msg_enable: %d\n",
2237                            net_device_ctx->msg_enable);
2238
2239         hv_set_drvdata(dev, net);
2240
2241         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
2242
2243         spin_lock_init(&net_device_ctx->lock);
2244         INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
2245         INIT_DELAYED_WORK(&net_device_ctx->vf_takeover, netvsc_vf_setup);
2246
2247         net_device_ctx->vf_stats
2248                 = netdev_alloc_pcpu_stats(struct netvsc_vf_pcpu_stats);
2249         if (!net_device_ctx->vf_stats)
2250                 goto no_stats;
2251
2252         net->netdev_ops = &device_ops;
2253         net->ethtool_ops = &ethtool_ops;
2254         SET_NETDEV_DEV(net, &dev->device);
2255
2256         /* We always need headroom for rndis header */
2257         net->needed_headroom = RNDIS_AND_PPI_SIZE;
2258
2259         /* Initialize the number of queues to be 1, we may change it if more
2260          * channels are offered later.
2261          */
2262         netif_set_real_num_tx_queues(net, 1);
2263         netif_set_real_num_rx_queues(net, 1);
2264
2265         /* Notify the netvsc driver of the new device */
2266         device_info = netvsc_devinfo_get(NULL);
2267
2268         if (!device_info) {
2269                 ret = -ENOMEM;
2270                 goto devinfo_failed;
2271         }
2272
2273         nvdev = rndis_filter_device_add(dev, device_info);
2274         if (IS_ERR(nvdev)) {
2275                 ret = PTR_ERR(nvdev);
2276                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
2277                 goto rndis_failed;
2278         }
2279
2280         memcpy(net->dev_addr, device_info->mac_adr, ETH_ALEN);
2281
2282         /* We must get rtnl lock before scheduling nvdev->subchan_work,
2283          * otherwise netvsc_subchan_work() can get rtnl lock first and wait
2284          * all subchannels to show up, but that may not happen because
2285          * netvsc_probe() can't get rtnl lock and as a result vmbus_onoffer()
2286          * -> ... -> device_add() -> ... -> __device_attach() can't get
2287          * the device lock, so all the subchannels can't be processed --
2288          * finally netvsc_subchan_work() hangs forever.
2289          */
2290         rtnl_lock();
2291
2292         if (nvdev->num_chn > 1)
2293                 schedule_work(&nvdev->subchan_work);
2294
2295         /* hw_features computed in rndis_netdev_set_hwcaps() */
2296         net->features = net->hw_features |
2297                 NETIF_F_HIGHDMA | NETIF_F_SG |
2298                 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
2299         net->vlan_features = net->features;
2300
2301         netdev_lockdep_set_classes(net);
2302
2303         /* MTU range: 68 - 1500 or 65521 */
2304         net->min_mtu = NETVSC_MTU_MIN;
2305         if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
2306                 net->max_mtu = NETVSC_MTU - ETH_HLEN;
2307         else
2308                 net->max_mtu = ETH_DATA_LEN;
2309
2310         ret = register_netdevice(net);
2311         if (ret != 0) {
2312                 pr_err("Unable to register netdev.\n");
2313                 goto register_failed;
2314         }
2315
2316         list_add(&net_device_ctx->list, &netvsc_dev_list);
2317         rtnl_unlock();
2318
2319         kfree(device_info);
2320         return 0;
2321
2322 register_failed:
2323         rtnl_unlock();
2324         rndis_filter_device_remove(dev, nvdev);
2325 rndis_failed:
2326         kfree(device_info);
2327 devinfo_failed:
2328         free_percpu(net_device_ctx->vf_stats);
2329 no_stats:
2330         hv_set_drvdata(dev, NULL);
2331         free_netdev(net);
2332 no_net:
2333         return ret;
2334 }
2335
2336 static int netvsc_remove(struct hv_device *dev)
2337 {
2338         struct net_device_context *ndev_ctx;
2339         struct net_device *vf_netdev, *net;
2340         struct netvsc_device *nvdev;
2341
2342         net = hv_get_drvdata(dev);
2343         if (net == NULL) {
2344                 dev_err(&dev->device, "No net device to remove\n");
2345                 return 0;
2346         }
2347
2348         ndev_ctx = netdev_priv(net);
2349
2350         cancel_delayed_work_sync(&ndev_ctx->dwork);
2351
2352         rtnl_lock();
2353         nvdev = rtnl_dereference(ndev_ctx->nvdev);
2354         if (nvdev)
2355                 cancel_work_sync(&nvdev->subchan_work);
2356
2357         /*
2358          * Call to the vsc driver to let it know that the device is being
2359          * removed. Also blocks mtu and channel changes.
2360          */
2361         vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2362         if (vf_netdev)
2363                 netvsc_unregister_vf(vf_netdev);
2364
2365         if (nvdev)
2366                 rndis_filter_device_remove(dev, nvdev);
2367
2368         unregister_netdevice(net);
2369         list_del(&ndev_ctx->list);
2370
2371         rtnl_unlock();
2372
2373         hv_set_drvdata(dev, NULL);
2374
2375         free_percpu(ndev_ctx->vf_stats);
2376         free_netdev(net);
2377         return 0;
2378 }
2379
2380 static const struct hv_vmbus_device_id id_table[] = {
2381         /* Network guid */
2382         { HV_NIC_GUID, },
2383         { },
2384 };
2385
2386 MODULE_DEVICE_TABLE(vmbus, id_table);
2387
2388 /* The one and only one */
2389 static struct  hv_driver netvsc_drv = {
2390         .name = KBUILD_MODNAME,
2391         .id_table = id_table,
2392         .probe = netvsc_probe,
2393         .remove = netvsc_remove,
2394         .driver = {
2395                 .probe_type = PROBE_PREFER_ASYNCHRONOUS,
2396         },
2397 };
2398
2399 /*
2400  * On Hyper-V, every VF interface is matched with a corresponding
2401  * synthetic interface. The synthetic interface is presented first
2402  * to the guest. When the corresponding VF instance is registered,
2403  * we will take care of switching the data path.
2404  */
2405 static int netvsc_netdev_event(struct notifier_block *this,
2406                                unsigned long event, void *ptr)
2407 {
2408         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2409
2410         /* Skip our own events */
2411         if (event_dev->netdev_ops == &device_ops)
2412                 return NOTIFY_DONE;
2413
2414         /* Avoid non-Ethernet type devices */
2415         if (event_dev->type != ARPHRD_ETHER)
2416                 return NOTIFY_DONE;
2417
2418         /* Avoid Vlan dev with same MAC registering as VF */
2419         if (is_vlan_dev(event_dev))
2420                 return NOTIFY_DONE;
2421
2422         /* Avoid Bonding master dev with same MAC registering as VF */
2423         if ((event_dev->priv_flags & IFF_BONDING) &&
2424             (event_dev->flags & IFF_MASTER))
2425                 return NOTIFY_DONE;
2426
2427         switch (event) {
2428         case NETDEV_REGISTER:
2429                 return netvsc_register_vf(event_dev);
2430         case NETDEV_UNREGISTER:
2431                 return netvsc_unregister_vf(event_dev);
2432         case NETDEV_UP:
2433         case NETDEV_DOWN:
2434                 return netvsc_vf_changed(event_dev);
2435         default:
2436                 return NOTIFY_DONE;
2437         }
2438 }
2439
2440 static struct notifier_block netvsc_netdev_notifier = {
2441         .notifier_call = netvsc_netdev_event,
2442 };
2443
2444 static void __exit netvsc_drv_exit(void)
2445 {
2446         unregister_netdevice_notifier(&netvsc_netdev_notifier);
2447         vmbus_driver_unregister(&netvsc_drv);
2448 }
2449
2450 static int __init netvsc_drv_init(void)
2451 {
2452         int ret;
2453
2454         if (ring_size < RING_SIZE_MIN) {
2455                 ring_size = RING_SIZE_MIN;
2456                 pr_info("Increased ring_size to %u (min allowed)\n",
2457                         ring_size);
2458         }
2459         netvsc_ring_bytes = ring_size * PAGE_SIZE;
2460
2461         ret = vmbus_driver_register(&netvsc_drv);
2462         if (ret)
2463                 return ret;
2464
2465         register_netdevice_notifier(&netvsc_netdev_notifier);
2466         return 0;
2467 }
2468
2469 MODULE_LICENSE("GPL");
2470 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
2471
2472 module_init(netvsc_drv_init);
2473 module_exit(netvsc_drv_exit);