drivers/net/hyperv/netvsc_drv.c

   1 /*
   2  * Copyright (c) 2009, Microsoft Corporation.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms and conditions of the GNU General Public License,
   6  * version 2, as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope it will be useful, but WITHOUT
   9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  11  * more details.
  12  *
  13  * You should have received a copy of the GNU General Public License along with
  14  * this program; if not, see <http://www.gnu.org/licenses/>.
  15  *
  16  * Authors:
  17  *   Haiyang Zhang <haiyangz@microsoft.com>
  18  *   Hank Janssen  <hjanssen@microsoft.com>
  19  */
  20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  21
  22 #include <linux/init.h>
  23 #include <linux/atomic.h>
  24 #include <linux/module.h>
  25 #include <linux/highmem.h>
  26 #include <linux/device.h>
  27 #include <linux/io.h>
  28 #include <linux/delay.h>
  29 #include <linux/netdevice.h>
  30 #include <linux/inetdevice.h>
  31 #include <linux/etherdevice.h>
  32 #include <linux/pci.h>
  33 #include <linux/skbuff.h>
  34 #include <linux/if_vlan.h>
  35 #include <linux/in.h>
  36 #include <linux/slab.h>
  37 #include <linux/rtnetlink.h>
  38 #include <linux/netpoll.h>
  39
  40 #include <net/arp.h>
  41 #include <net/route.h>
  42 #include <net/sock.h>
  43 #include <net/pkt_sched.h>
  44 #include <net/checksum.h>
  45 #include <net/ip6_checksum.h>
  46
  47 #include "hyperv_net.h"
  48
  49 #define RING_SIZE_MIN   64
  50 #define RETRY_US_LO     5000
  51 #define RETRY_US_HI     10000
  52 #define RETRY_MAX       2000    /* >10 sec */
  53
  54 #define LINKCHANGE_INT (2 * HZ)
  55 #define VF_TAKEOVER_INT (HZ / 10)
  56
  57 static unsigned int ring_size __ro_after_init = 128;
  58 module_param(ring_size, uint, 0444);
  59 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
  60 unsigned int netvsc_ring_bytes __ro_after_init;
  61
  62 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
  63                                 NETIF_MSG_LINK | NETIF_MSG_IFUP |
  64                                 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
  65                                 NETIF_MSG_TX_ERR;
  66
  67 static int debug = -1;
  68 module_param(debug, int, 0444);
  69 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  70
  71 static LIST_HEAD(netvsc_dev_list);
  72
  73 static void netvsc_change_rx_flags(struct net_device *net, int change)
  74 {
  75         struct net_device_context *ndev_ctx = netdev_priv(net);
  76         struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
  77         int inc;
  78
  79         if (!vf_netdev)
  80                 return;
  81
  82         if (change & IFF_PROMISC) {
  83                 inc = (net->flags & IFF_PROMISC) ? 1 : -1;
  84                 dev_set_promiscuity(vf_netdev, inc);
  85         }
  86
  87         if (change & IFF_ALLMULTI) {
  88                 inc = (net->flags & IFF_ALLMULTI) ? 1 : -1;
  89                 dev_set_allmulti(vf_netdev, inc);
  90         }
  91 }
  92
  93 static void netvsc_set_rx_mode(struct net_device *net)
  94 {
  95         struct net_device_context *ndev_ctx = netdev_priv(net);
  96         struct net_device *vf_netdev;
  97         struct netvsc_device *nvdev;
  98
  99         rcu_read_lock();
 100         vf_netdev = rcu_dereference(ndev_ctx->vf_netdev);
 101         if (vf_netdev) {
 102                 dev_uc_sync(vf_netdev, net);
 103                 dev_mc_sync(vf_netdev, net);
 104         }
 105
 106         nvdev = rcu_dereference(ndev_ctx->nvdev);
 107         if (nvdev)
 108                 rndis_filter_update(nvdev);
 109         rcu_read_unlock();
 110 }
 111
 112 static int netvsc_open(struct net_device *net)
 113 {
 114         struct net_device_context *ndev_ctx = netdev_priv(net);
 115         struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
 116         struct netvsc_device *nvdev = rtnl_dereference(ndev_ctx->nvdev);
 117         struct rndis_device *rdev;
 118         int ret = 0;
 119
 120         netif_carrier_off(net);
 121
 122         /* Open up the device */
 123         ret = rndis_filter_open(nvdev);
 124         if (ret != 0) {
 125                 netdev_err(net, "unable to open device (ret %d).\n", ret);
 126                 return ret;
 127         }
 128
 129         rdev = nvdev->extension;
 130         if (!rdev->link_state) {
 131                 netif_carrier_on(net);
 132                 netif_tx_wake_all_queues(net);
 133         }
 134
 135         if (vf_netdev) {
 136                 /* Setting synthetic device up transparently sets
 137                  * slave as up. If open fails, then slave will be
 138                  * still be offline (and not used).
 139                  */
 140                 ret = dev_open(vf_netdev);
 141                 if (ret)
 142                         netdev_warn(net,
 143                                     "unable to open slave: %s: %d\n",
 144                                     vf_netdev->name, ret);
 145         }
 146         return 0;
 147 }
 148
 149 static int netvsc_wait_until_empty(struct netvsc_device *nvdev)
 150 {
 151         unsigned int retry = 0;
 152         int i;
 153
 154         /* Ensure pending bytes in ring are read */
 155         for (;;) {
 156                 u32 aread = 0;
 157
 158                 for (i = 0; i < nvdev->num_chn; i++) {
 159                         struct vmbus_channel *chn
 160                                 = nvdev->chan_table[i].channel;
 161
 162                         if (!chn)
 163                                 continue;
 164
 165                         /* make sure receive not running now */
 166                         napi_synchronize(&nvdev->chan_table[i].napi);
 167
 168                         aread = hv_get_bytes_to_read(&chn->inbound);
 169                         if (aread)
 170                                 break;
 171
 172                         aread = hv_get_bytes_to_read(&chn->outbound);
 173                         if (aread)
 174                                 break;
 175                 }
 176
 177                 if (aread == 0)
 178                         return 0;
 179
 180                 if (++retry > RETRY_MAX)
 181                         return -ETIMEDOUT;
 182
 183                 usleep_range(RETRY_US_LO, RETRY_US_HI);
 184         }
 185 }
 186
 187 static int netvsc_close(struct net_device *net)
 188 {
 189         struct net_device_context *net_device_ctx = netdev_priv(net);
 190         struct net_device *vf_netdev
 191                 = rtnl_dereference(net_device_ctx->vf_netdev);
 192         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
 193         int ret;
 194
 195         netif_tx_disable(net);
 196
 197         /* No need to close rndis filter if it is removed already */
 198         if (!nvdev)
 199                 return 0;
 200
 201         ret = rndis_filter_close(nvdev);
 202         if (ret != 0) {
 203                 netdev_err(net, "unable to close device (ret %d).\n", ret);
 204                 return ret;
 205         }
 206
 207         ret = netvsc_wait_until_empty(nvdev);
 208         if (ret)
 209                 netdev_err(net, "Ring buffer not empty after closing rndis\n");
 210
 211         if (vf_netdev)
 212                 dev_close(vf_netdev);
 213
 214         return ret;
 215 }
 216
 217 static inline void *init_ppi_data(struct rndis_message *msg,
 218                                   u32 ppi_size, u32 pkt_type)
 219 {
 220         struct rndis_packet *rndis_pkt = &msg->msg.pkt;
 221         struct rndis_per_packet_info *ppi;
 222
 223         rndis_pkt->data_offset += ppi_size;
 224         ppi = (void *)rndis_pkt + rndis_pkt->per_pkt_info_offset
 225                 + rndis_pkt->per_pkt_info_len;
 226
 227         ppi->size = ppi_size;
 228         ppi->type = pkt_type;
 229         ppi->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                                select_queue_fallback_t fallback)
 336 {
 337         struct net_device_context *ndc = netdev_priv(ndev);
 338         struct net_device *vf_netdev;
 339         u16 txq;
 340
 341         rcu_read_lock();
 342         vf_netdev = rcu_dereference(ndc->vf_netdev);
 343         if (vf_netdev) {
 344                 const struct net_device_ops *vf_ops = vf_netdev->netdev_ops;
 345
 346                 if (vf_ops->ndo_select_queue)
 347                         txq = vf_ops->ndo_select_queue(vf_netdev, skb,
 348                                                        sb_dev, fallback);
 349                 else
 350                         txq = fallback(vf_netdev, skb, NULL);
 351
 352                 /* Record the queue selected by VF so that it can be
 353                  * used for common case where VF has more queues than
 354                  * the synthetic device.
 355                  */
 356                 qdisc_skb_cb(skb)->slave_dev_queue_mapping = txq;
 357         } else {
 358                 txq = netvsc_pick_tx(ndev, skb);
 359         }
 360         rcu_read_unlock();
 361
 362         while (unlikely(txq >= ndev->real_num_tx_queues))
 363                 txq -= ndev->real_num_tx_queues;
 364
 365         return txq;
 366 }
 367
 368 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
 369                        struct hv_page_buffer *pb)
 370 {
 371         int j = 0;
 372
 373         /* Deal with compund pages by ignoring unused part
 374          * of the page.
 375          */
 376         page += (offset >> PAGE_SHIFT);
 377         offset &= ~PAGE_MASK;
 378
 379         while (len > 0) {
 380                 unsigned long bytes;
 381
 382                 bytes = PAGE_SIZE - offset;
 383                 if (bytes > len)
 384                         bytes = len;
 385                 pb[j].pfn = page_to_pfn(page);
 386                 pb[j].offset = offset;
 387                 pb[j].len = bytes;
 388
 389                 offset += bytes;
 390                 len -= bytes;
 391
 392                 if (offset == PAGE_SIZE && len) {
 393                         page++;
 394                         offset = 0;
 395                         j++;
 396                 }
 397         }
 398
 399         return j + 1;
 400 }
 401
 402 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
 403                            struct hv_netvsc_packet *packet,
 404                            struct hv_page_buffer *pb)
 405 {
 406         u32 slots_used = 0;
 407         char *data = skb->data;
 408         int frags = skb_shinfo(skb)->nr_frags;
 409         int i;
 410
 411         /* The packet is laid out thus:
 412          * 1. hdr: RNDIS header and PPI
 413          * 2. skb linear data
 414          * 3. skb fragment data
 415          */
 416         slots_used += fill_pg_buf(virt_to_page(hdr),
 417                                   offset_in_page(hdr),
 418                                   len, &pb[slots_used]);
 419
 420         packet->rmsg_size = len;
 421         packet->rmsg_pgcnt = slots_used;
 422
 423         slots_used += fill_pg_buf(virt_to_page(data),
 424                                 offset_in_page(data),
 425                                 skb_headlen(skb), &pb[slots_used]);
 426
 427         for (i = 0; i < frags; i++) {
 428                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 429
 430                 slots_used += fill_pg_buf(skb_frag_page(frag),
 431                                         frag->page_offset,
 432                                         skb_frag_size(frag), &pb[slots_used]);
 433         }
 434         return slots_used;
 435 }
 436
 437 static int count_skb_frag_slots(struct sk_buff *skb)
 438 {
 439         int i, frags = skb_shinfo(skb)->nr_frags;
 440         int pages = 0;
 441
 442         for (i = 0; i < frags; i++) {
 443                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 444                 unsigned long size = skb_frag_size(frag);
 445                 unsigned long offset = frag->page_offset;
 446
 447                 /* Skip unused frames from start of page */
 448                 offset &= ~PAGE_MASK;
 449                 pages += PFN_UP(offset + size);
 450         }
 451         return pages;
 452 }
 453
 454 static int netvsc_get_slots(struct sk_buff *skb)
 455 {
 456         char *data = skb->data;
 457         unsigned int offset = offset_in_page(data);
 458         unsigned int len = skb_headlen(skb);
 459         int slots;
 460         int frag_slots;
 461
 462         slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
 463         frag_slots = count_skb_frag_slots(skb);
 464         return slots + frag_slots;
 465 }
 466
 467 static u32 net_checksum_info(struct sk_buff *skb)
 468 {
 469         if (skb->protocol == htons(ETH_P_IP)) {
 470                 struct iphdr *ip = ip_hdr(skb);
 471
 472                 if (ip->protocol == IPPROTO_TCP)
 473                         return TRANSPORT_INFO_IPV4_TCP;
 474                 else if (ip->protocol == IPPROTO_UDP)
 475                         return TRANSPORT_INFO_IPV4_UDP;
 476         } else {
 477                 struct ipv6hdr *ip6 = ipv6_hdr(skb);
 478
 479                 if (ip6->nexthdr == IPPROTO_TCP)
 480                         return TRANSPORT_INFO_IPV6_TCP;
 481                 else if (ip6->nexthdr == IPPROTO_UDP)
 482                         return TRANSPORT_INFO_IPV6_UDP;
 483         }
 484
 485         return TRANSPORT_INFO_NOT_IP;
 486 }
 487
 488 /* Send skb on the slave VF device. */
 489 static int netvsc_vf_xmit(struct net_device *net, struct net_device *vf_netdev,
 490                           struct sk_buff *skb)
 491 {
 492         struct net_device_context *ndev_ctx = netdev_priv(net);
 493         unsigned int len = skb->len;
 494         int rc;
 495
 496         skb->dev = vf_netdev;
 497         skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
 498
 499         rc = dev_queue_xmit(skb);
 500         if (likely(rc == NET_XMIT_SUCCESS || rc == NET_XMIT_CN)) {
 501                 struct netvsc_vf_pcpu_stats *pcpu_stats
 502                         = this_cpu_ptr(ndev_ctx->vf_stats);
 503
 504                 u64_stats_update_begin(&pcpu_stats->syncp);
 505                 pcpu_stats->tx_packets++;
 506                 pcpu_stats->tx_bytes += len;
 507                 u64_stats_update_end(&pcpu_stats->syncp);
 508         } else {
 509                 this_cpu_inc(ndev_ctx->vf_stats->tx_dropped);
 510         }
 511
 512         return rc;
 513 }
 514
 515 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
 516 {
 517         struct net_device_context *net_device_ctx = netdev_priv(net);
 518         struct hv_netvsc_packet *packet = NULL;
 519         int ret;
 520         unsigned int num_data_pgs;
 521         struct rndis_message *rndis_msg;
 522         struct net_device *vf_netdev;
 523         u32 rndis_msg_size;
 524         u32 hash;
 525         struct hv_page_buffer pb[MAX_PAGE_BUFFER_COUNT];
 526
 527         /* if VF is present and up then redirect packets
 528          * already called with rcu_read_lock_bh
 529          */
 530         vf_netdev = rcu_dereference_bh(net_device_ctx->vf_netdev);
 531         if (vf_netdev && netif_running(vf_netdev) &&
 532             !netpoll_tx_running(net))
 533                 return netvsc_vf_xmit(net, vf_netdev, skb);
 534
 535         /* We will atmost need two pages to describe the rndis
 536          * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
 537          * of pages in a single packet. If skb is scattered around
 538          * more pages we try linearizing it.
 539          */
 540
 541         num_data_pgs = netvsc_get_slots(skb) + 2;
 542
 543         if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
 544                 ++net_device_ctx->eth_stats.tx_scattered;
 545
 546                 if (skb_linearize(skb))
 547                         goto no_memory;
 548
 549                 num_data_pgs = netvsc_get_slots(skb) + 2;
 550                 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
 551                         ++net_device_ctx->eth_stats.tx_too_big;
 552                         goto drop;
 553                 }
 554         }
 555
 556         /*
 557          * Place the rndis header in the skb head room and
 558          * the skb->cb will be used for hv_netvsc_packet
 559          * structure.
 560          */
 561         ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
 562         if (ret)
 563                 goto no_memory;
 564
 565         /* Use the skb control buffer for building up the packet */
 566         BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
 567                         FIELD_SIZEOF(struct sk_buff, cb));
 568         packet = (struct hv_netvsc_packet *)skb->cb;
 569
 570         packet->q_idx = skb_get_queue_mapping(skb);
 571
 572         packet->total_data_buflen = skb->len;
 573         packet->total_bytes = skb->len;
 574         packet->total_packets = 1;
 575
 576         rndis_msg = (struct rndis_message *)skb->head;
 577
 578         /* Add the rndis header */
 579         rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
 580         rndis_msg->msg_len = packet->total_data_buflen;
 581
 582         rndis_msg->msg.pkt = (struct rndis_packet) {
 583                 .data_offset = sizeof(struct rndis_packet),
 584                 .data_len = packet->total_data_buflen,
 585                 .per_pkt_info_offset = sizeof(struct rndis_packet),
 586         };
 587
 588         rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
 589
 590         hash = skb_get_hash_raw(skb);
 591         if (hash != 0 && net->real_num_tx_queues > 1) {
 592                 u32 *hash_info;
 593
 594                 rndis_msg_size += NDIS_HASH_PPI_SIZE;
 595                 hash_info = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
 596                                           NBL_HASH_VALUE);
 597                 *hash_info = hash;
 598         }
 599
 600         if (skb_vlan_tag_present(skb)) {
 601                 struct ndis_pkt_8021q_info *vlan;
 602
 603                 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
 604                 vlan = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
 605                                      IEEE_8021Q_INFO);
 606
 607                 vlan->value = 0;
 608                 vlan->vlanid = skb->vlan_tci & VLAN_VID_MASK;
 609                 vlan->pri = (skb->vlan_tci & VLAN_PRIO_MASK) >>
 610                                 VLAN_PRIO_SHIFT;
 611         }
 612
 613         if (skb_is_gso(skb)) {
 614                 struct ndis_tcp_lso_info *lso_info;
 615
 616                 rndis_msg_size += NDIS_LSO_PPI_SIZE;
 617                 lso_info = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
 618                                          TCP_LARGESEND_PKTINFO);
 619
 620                 lso_info->value = 0;
 621                 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
 622                 if (skb->protocol == htons(ETH_P_IP)) {
 623                         lso_info->lso_v2_transmit.ip_version =
 624                                 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
 625                         ip_hdr(skb)->tot_len = 0;
 626                         ip_hdr(skb)->check = 0;
 627                         tcp_hdr(skb)->check =
 628                                 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
 629                                                    ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 630                 } else {
 631                         lso_info->lso_v2_transmit.ip_version =
 632                                 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
 633                         ipv6_hdr(skb)->payload_len = 0;
 634                         tcp_hdr(skb)->check =
 635                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 636                                                  &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 637                 }
 638                 lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb);
 639                 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
 640         } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
 641                 if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) {
 642                         struct ndis_tcp_ip_checksum_info *csum_info;
 643
 644                         rndis_msg_size += NDIS_CSUM_PPI_SIZE;
 645                         csum_info = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
 646                                                   TCPIP_CHKSUM_PKTINFO);
 647
 648                         csum_info->value = 0;
 649                         csum_info->transmit.tcp_header_offset = skb_transport_offset(skb);
 650
 651                         if (skb->protocol == htons(ETH_P_IP)) {
 652                                 csum_info->transmit.is_ipv4 = 1;
 653
 654                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
 655                                         csum_info->transmit.tcp_checksum = 1;
 656                                 else
 657                                         csum_info->transmit.udp_checksum = 1;
 658                         } else {
 659                                 csum_info->transmit.is_ipv6 = 1;
 660
 661                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
 662                                         csum_info->transmit.tcp_checksum = 1;
 663                                 else
 664                                         csum_info->transmit.udp_checksum = 1;
 665                         }
 666                 } else {
 667                         /* Can't do offload of this type of checksum */
 668                         if (skb_checksum_help(skb))
 669                                 goto drop;
 670                 }
 671         }
 672
 673         /* Start filling in the page buffers with the rndis hdr */
 674         rndis_msg->msg_len += rndis_msg_size;
 675         packet->total_data_buflen = rndis_msg->msg_len;
 676         packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
 677                                                skb, packet, pb);
 678
 679         /* timestamp packet in software */
 680         skb_tx_timestamp(skb);
 681
 682         ret = netvsc_send(net, packet, rndis_msg, pb, skb);
 683         if (likely(ret == 0))
 684                 return NETDEV_TX_OK;
 685
 686         if (ret == -EAGAIN) {
 687                 ++net_device_ctx->eth_stats.tx_busy;
 688                 return NETDEV_TX_BUSY;
 689         }
 690
 691         if (ret == -ENOSPC)
 692                 ++net_device_ctx->eth_stats.tx_no_space;
 693
 694 drop:
 695         dev_kfree_skb_any(skb);
 696         net->stats.tx_dropped++;
 697
 698         return NETDEV_TX_OK;
 699
 700 no_memory:
 701         ++net_device_ctx->eth_stats.tx_no_memory;
 702         goto drop;
 703 }
 704
 705 /*
 706  * netvsc_linkstatus_callback - Link up/down notification
 707  */
 708 void netvsc_linkstatus_callback(struct net_device *net,
 709                                 struct rndis_message *resp)
 710 {
 711         struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
 712         struct net_device_context *ndev_ctx = netdev_priv(net);
 713         struct netvsc_reconfig *event;
 714         unsigned long flags;
 715
 716         /* Update the physical link speed when changing to another vSwitch */
 717         if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
 718                 u32 speed;
 719
 720                 speed = *(u32 *)((void *)indicate
 721                                  + indicate->status_buf_offset) / 10000;
 722                 ndev_ctx->speed = speed;
 723                 return;
 724         }
 725
 726         /* Handle these link change statuses below */
 727         if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
 728             indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
 729             indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
 730                 return;
 731
 732         if (net->reg_state != NETREG_REGISTERED)
 733                 return;
 734
 735         event = kzalloc(sizeof(*event), GFP_ATOMIC);
 736         if (!event)
 737                 return;
 738         event->event = indicate->status;
 739
 740         spin_lock_irqsave(&ndev_ctx->lock, flags);
 741         list_add_tail(&event->list, &ndev_ctx->reconfig_events);
 742         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
 743
 744         schedule_delayed_work(&ndev_ctx->dwork, 0);
 745 }
 746
 747 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
 748                                              struct netvsc_channel *nvchan)
 749 {
 750         struct napi_struct *napi = &nvchan->napi;
 751         const struct ndis_pkt_8021q_info *vlan = nvchan->rsc.vlan;
 752         const struct ndis_tcp_ip_checksum_info *csum_info =
 753                                                 nvchan->rsc.csum_info;
 754         struct sk_buff *skb;
 755         int i;
 756
 757         skb = napi_alloc_skb(napi, nvchan->rsc.pktlen);
 758         if (!skb)
 759                 return skb;
 760
 761         /*
 762          * Copy to skb. This copy is needed here since the memory pointed by
 763          * hv_netvsc_packet cannot be deallocated
 764          */
 765         for (i = 0; i < nvchan->rsc.cnt; i++)
 766                 skb_put_data(skb, nvchan->rsc.data[i], nvchan->rsc.len[i]);
 767
 768         skb->protocol = eth_type_trans(skb, net);
 769
 770         /* skb is already created with CHECKSUM_NONE */
 771         skb_checksum_none_assert(skb);
 772
 773         /*
 774          * In Linux, the IP checksum is always checked.
 775          * Do L4 checksum offload if enabled and present.
 776          */
 777         if (csum_info && (net->features & NETIF_F_RXCSUM)) {
 778                 if (csum_info->receive.tcp_checksum_succeeded ||
 779                     csum_info->receive.udp_checksum_succeeded)
 780                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 781         }
 782
 783         if (vlan) {
 784                 u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT);
 785
 786                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 787                                        vlan_tci);
 788         }
 789
 790         return skb;
 791 }
 792
 793 /*
 794  * netvsc_recv_callback -  Callback when we receive a packet from the
 795  * "wire" on the specified device.
 796  */
 797 int netvsc_recv_callback(struct net_device *net,
 798                          struct netvsc_device *net_device,
 799                          struct netvsc_channel *nvchan)
 800 {
 801         struct net_device_context *net_device_ctx = netdev_priv(net);
 802         struct vmbus_channel *channel = nvchan->channel;
 803         u16 q_idx = channel->offermsg.offer.sub_channel_index;
 804         struct sk_buff *skb;
 805         struct netvsc_stats *rx_stats;
 806
 807         if (net->reg_state != NETREG_REGISTERED)
 808                 return NVSP_STAT_FAIL;
 809
 810         /* Allocate a skb - TODO direct I/O to pages? */
 811         skb = netvsc_alloc_recv_skb(net, nvchan);
 812
 813         if (unlikely(!skb)) {
 814                 ++net_device_ctx->eth_stats.rx_no_memory;
 815                 rcu_read_unlock();
 816                 return NVSP_STAT_FAIL;
 817         }
 818
 819         skb_record_rx_queue(skb, q_idx);
 820
 821         /*
 822          * Even if injecting the packet, record the statistics
 823          * on the synthetic device because modifying the VF device
 824          * statistics will not work correctly.
 825          */
 826         rx_stats = &nvchan->rx_stats;
 827         u64_stats_update_begin(&rx_stats->syncp);
 828         rx_stats->packets++;
 829         rx_stats->bytes += nvchan->rsc.pktlen;
 830
 831         if (skb->pkt_type == PACKET_BROADCAST)
 832                 ++rx_stats->broadcast;
 833         else if (skb->pkt_type == PACKET_MULTICAST)
 834                 ++rx_stats->multicast;
 835         u64_stats_update_end(&rx_stats->syncp);
 836
 837         napi_gro_receive(&nvchan->napi, skb);
 838         return NVSP_STAT_SUCCESS;
 839 }
 840
 841 static void netvsc_get_drvinfo(struct net_device *net,
 842                                struct ethtool_drvinfo *info)
 843 {
 844         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
 845         strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
 846 }
 847
 848 static void netvsc_get_channels(struct net_device *net,
 849                                 struct ethtool_channels *channel)
 850 {
 851         struct net_device_context *net_device_ctx = netdev_priv(net);
 852         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
 853
 854         if (nvdev) {
 855                 channel->max_combined   = nvdev->max_chn;
 856                 channel->combined_count = nvdev->num_chn;
 857         }
 858 }
 859
 860 static int netvsc_detach(struct net_device *ndev,
 861                          struct netvsc_device *nvdev)
 862 {
 863         struct net_device_context *ndev_ctx = netdev_priv(ndev);
 864         struct hv_device *hdev = ndev_ctx->device_ctx;
 865         int ret;
 866
 867         /* Don't try continuing to try and setup sub channels */
 868         if (cancel_work_sync(&nvdev->subchan_work))
 869                 nvdev->num_chn = 1;
 870
 871         /* If device was up (receiving) then shutdown */
 872         if (netif_running(ndev)) {
 873                 netif_tx_disable(ndev);
 874
 875                 ret = rndis_filter_close(nvdev);
 876                 if (ret) {
 877                         netdev_err(ndev,
 878                                    "unable to close device (ret %d).\n", ret);
 879                         return ret;
 880                 }
 881
 882                 ret = netvsc_wait_until_empty(nvdev);
 883                 if (ret) {
 884                         netdev_err(ndev,
 885                                    "Ring buffer not empty after closing rndis\n");
 886                         return ret;
 887                 }
 888         }
 889
 890         netif_device_detach(ndev);
 891
 892         rndis_filter_device_remove(hdev, nvdev);
 893
 894         return 0;
 895 }
 896
 897 static int netvsc_attach(struct net_device *ndev,
 898                          struct netvsc_device_info *dev_info)
 899 {
 900         struct net_device_context *ndev_ctx = netdev_priv(ndev);
 901         struct hv_device *hdev = ndev_ctx->device_ctx;
 902         struct netvsc_device *nvdev;
 903         struct rndis_device *rdev;
 904         int ret;
 905
 906         nvdev = rndis_filter_device_add(hdev, dev_info);
 907         if (IS_ERR(nvdev))
 908                 return PTR_ERR(nvdev);
 909
 910         if (nvdev->num_chn > 1) {
 911                 ret = rndis_set_subchannel(ndev, nvdev);
 912
 913                 /* if unavailable, just proceed with one queue */
 914                 if (ret) {
 915                         nvdev->max_chn = 1;
 916                         nvdev->num_chn = 1;
 917                 }
 918         }
 919
 920         /* In any case device is now ready */
 921         netif_device_attach(ndev);
 922
 923         /* Note: enable and attach happen when sub-channels setup */
 924         netif_carrier_off(ndev);
 925
 926         if (netif_running(ndev)) {
 927                 ret = rndis_filter_open(nvdev);
 928                 if (ret)
 929                         return ret;
 930
 931                 rdev = nvdev->extension;
 932                 if (!rdev->link_state)
 933                         netif_carrier_on(ndev);
 934         }
 935
 936         return 0;
 937 }
 938
 939 static int netvsc_set_channels(struct net_device *net,
 940                                struct ethtool_channels *channels)
 941 {
 942         struct net_device_context *net_device_ctx = netdev_priv(net);
 943         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
 944         unsigned int orig, count = channels->combined_count;
 945         struct netvsc_device_info device_info;
 946         int ret;
 947
 948         /* We do not support separate count for rx, tx, or other */
 949         if (count == 0 ||
 950             channels->rx_count || channels->tx_count || channels->other_count)
 951                 return -EINVAL;
 952
 953         if (!nvdev || nvdev->destroy)
 954                 return -ENODEV;
 955
 956         if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
 957                 return -EINVAL;
 958
 959         if (count > nvdev->max_chn)
 960                 return -EINVAL;
 961
 962         orig = nvdev->num_chn;
 963
 964         memset(&device_info, 0, sizeof(device_info));
 965         device_info.num_chn = count;
 966         device_info.send_sections = nvdev->send_section_cnt;
 967         device_info.send_section_size = nvdev->send_section_size;
 968         device_info.recv_sections = nvdev->recv_section_cnt;
 969         device_info.recv_section_size = nvdev->recv_section_size;
 970
 971         ret = netvsc_detach(net, nvdev);
 972         if (ret)
 973                 return ret;
 974
 975         ret = netvsc_attach(net, &device_info);
 976         if (ret) {
 977                 device_info.num_chn = orig;
 978                 if (netvsc_attach(net, &device_info))
 979                         netdev_err(net, "restoring channel setting failed\n");
 980         }
 981
 982         return ret;
 983 }
 984
 985 static bool
 986 netvsc_validate_ethtool_ss_cmd(const struct ethtool_link_ksettings *cmd)
 987 {
 988         struct ethtool_link_ksettings diff1 = *cmd;
 989         struct ethtool_link_ksettings diff2 = {};
 990
 991         diff1.base.speed = 0;
 992         diff1.base.duplex = 0;
 993         /* advertising and cmd are usually set */
 994         ethtool_link_ksettings_zero_link_mode(&diff1, advertising);
 995         diff1.base.cmd = 0;
 996         /* We set port to PORT_OTHER */
 997         diff2.base.port = PORT_OTHER;
 998
 999         return !memcmp(&diff1, &diff2, sizeof(diff1));
1000 }
1001
1002 static void netvsc_init_settings(struct net_device *dev)
1003 {
1004         struct net_device_context *ndc = netdev_priv(dev);
1005
1006         ndc->l4_hash = HV_DEFAULT_L4HASH;
1007
1008         ndc->speed = SPEED_UNKNOWN;
1009         ndc->duplex = DUPLEX_FULL;
1010
1011         dev->features = NETIF_F_LRO;
1012 }
1013
1014 static int netvsc_get_link_ksettings(struct net_device *dev,
1015                                      struct ethtool_link_ksettings *cmd)
1016 {
1017         struct net_device_context *ndc = netdev_priv(dev);
1018
1019         cmd->base.speed = ndc->speed;
1020         cmd->base.duplex = ndc->duplex;
1021         cmd->base.port = PORT_OTHER;
1022
1023         return 0;
1024 }
1025
1026 static int netvsc_set_link_ksettings(struct net_device *dev,
1027                                      const struct ethtool_link_ksettings *cmd)
1028 {
1029         struct net_device_context *ndc = netdev_priv(dev);
1030         u32 speed;
1031
1032         speed = cmd->base.speed;
1033         if (!ethtool_validate_speed(speed) ||
1034             !ethtool_validate_duplex(cmd->base.duplex) ||
1035             !netvsc_validate_ethtool_ss_cmd(cmd))
1036                 return -EINVAL;
1037
1038         ndc->speed = speed;
1039         ndc->duplex = cmd->base.duplex;
1040
1041         return 0;
1042 }
1043
1044 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
1045 {
1046         struct net_device_context *ndevctx = netdev_priv(ndev);
1047         struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1048         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1049         int orig_mtu = ndev->mtu;
1050         struct netvsc_device_info device_info;
1051         int ret = 0;
1052
1053         if (!nvdev || nvdev->destroy)
1054                 return -ENODEV;
1055
1056         /* Change MTU of underlying VF netdev first. */
1057         if (vf_netdev) {
1058                 ret = dev_set_mtu(vf_netdev, mtu);
1059                 if (ret)
1060                         return ret;
1061         }
1062
1063         memset(&device_info, 0, sizeof(device_info));
1064         device_info.num_chn = nvdev->num_chn;
1065         device_info.send_sections = nvdev->send_section_cnt;
1066         device_info.send_section_size = nvdev->send_section_size;
1067         device_info.recv_sections = nvdev->recv_section_cnt;
1068         device_info.recv_section_size = nvdev->recv_section_size;
1069
1070         ret = netvsc_detach(ndev, nvdev);
1071         if (ret)
1072                 goto rollback_vf;
1073
1074         ndev->mtu = mtu;
1075
1076         ret = netvsc_attach(ndev, &device_info);
1077         if (ret)
1078                 goto rollback;
1079
1080         return 0;
1081
1082 rollback:
1083         /* Attempt rollback to original MTU */
1084         ndev->mtu = orig_mtu;
1085
1086         if (netvsc_attach(ndev, &device_info))
1087                 netdev_err(ndev, "restoring mtu failed\n");
1088 rollback_vf:
1089         if (vf_netdev)
1090                 dev_set_mtu(vf_netdev, orig_mtu);
1091
1092         return ret;
1093 }
1094
1095 static void netvsc_get_vf_stats(struct net_device *net,
1096                                 struct netvsc_vf_pcpu_stats *tot)
1097 {
1098         struct net_device_context *ndev_ctx = netdev_priv(net);
1099         int i;
1100
1101         memset(tot, 0, sizeof(*tot));
1102
1103         for_each_possible_cpu(i) {
1104                 const struct netvsc_vf_pcpu_stats *stats
1105                         = per_cpu_ptr(ndev_ctx->vf_stats, i);
1106                 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1107                 unsigned int start;
1108
1109                 do {
1110                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1111                         rx_packets = stats->rx_packets;
1112                         tx_packets = stats->tx_packets;
1113                         rx_bytes = stats->rx_bytes;
1114                         tx_bytes = stats->tx_bytes;
1115                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1116
1117                 tot->rx_packets += rx_packets;
1118                 tot->tx_packets += tx_packets;
1119                 tot->rx_bytes   += rx_bytes;
1120                 tot->tx_bytes   += tx_bytes;
1121                 tot->tx_dropped += stats->tx_dropped;
1122         }
1123 }
1124
1125 static void netvsc_get_pcpu_stats(struct net_device *net,
1126                                   struct netvsc_ethtool_pcpu_stats *pcpu_tot)
1127 {
1128         struct net_device_context *ndev_ctx = netdev_priv(net);
1129         struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
1130         int i;
1131
1132         /* fetch percpu stats of vf */
1133         for_each_possible_cpu(i) {
1134                 const struct netvsc_vf_pcpu_stats *stats =
1135                         per_cpu_ptr(ndev_ctx->vf_stats, i);
1136                 struct netvsc_ethtool_pcpu_stats *this_tot = &pcpu_tot[i];
1137                 unsigned int start;
1138
1139                 do {
1140                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1141                         this_tot->vf_rx_packets = stats->rx_packets;
1142                         this_tot->vf_tx_packets = stats->tx_packets;
1143                         this_tot->vf_rx_bytes = stats->rx_bytes;
1144                         this_tot->vf_tx_bytes = stats->tx_bytes;
1145                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1146                 this_tot->rx_packets = this_tot->vf_rx_packets;
1147                 this_tot->tx_packets = this_tot->vf_tx_packets;
1148                 this_tot->rx_bytes   = this_tot->vf_rx_bytes;
1149                 this_tot->tx_bytes   = this_tot->vf_tx_bytes;
1150         }
1151
1152         /* fetch percpu stats of netvsc */
1153         for (i = 0; i < nvdev->num_chn; i++) {
1154                 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1155                 const struct netvsc_stats *stats;
1156                 struct netvsc_ethtool_pcpu_stats *this_tot =
1157                         &pcpu_tot[nvchan->channel->target_cpu];
1158                 u64 packets, bytes;
1159                 unsigned int start;
1160
1161                 stats = &nvchan->tx_stats;
1162                 do {
1163                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1164                         packets = stats->packets;
1165                         bytes = stats->bytes;
1166                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1167
1168                 this_tot->tx_bytes      += bytes;
1169                 this_tot->tx_packets    += packets;
1170
1171                 stats = &nvchan->rx_stats;
1172                 do {
1173                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1174                         packets = stats->packets;
1175                         bytes = stats->bytes;
1176                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1177
1178                 this_tot->rx_bytes      += bytes;
1179                 this_tot->rx_packets    += packets;
1180         }
1181 }
1182
1183 static void netvsc_get_stats64(struct net_device *net,
1184                                struct rtnl_link_stats64 *t)
1185 {
1186         struct net_device_context *ndev_ctx = netdev_priv(net);
1187         struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
1188         struct netvsc_vf_pcpu_stats vf_tot;
1189         int i;
1190
1191         if (!nvdev)
1192                 return;
1193
1194         netdev_stats_to_stats64(t, &net->stats);
1195
1196         netvsc_get_vf_stats(net, &vf_tot);
1197         t->rx_packets += vf_tot.rx_packets;
1198         t->tx_packets += vf_tot.tx_packets;
1199         t->rx_bytes   += vf_tot.rx_bytes;
1200         t->tx_bytes   += vf_tot.tx_bytes;
1201         t->tx_dropped += vf_tot.tx_dropped;
1202
1203         for (i = 0; i < nvdev->num_chn; i++) {
1204                 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1205                 const struct netvsc_stats *stats;
1206                 u64 packets, bytes, multicast;
1207                 unsigned int start;
1208
1209                 stats = &nvchan->tx_stats;
1210                 do {
1211                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1212                         packets = stats->packets;
1213                         bytes = stats->bytes;
1214                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1215
1216                 t->tx_bytes     += bytes;
1217                 t->tx_packets   += packets;
1218
1219                 stats = &nvchan->rx_stats;
1220                 do {
1221                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1222                         packets = stats->packets;
1223                         bytes = stats->bytes;
1224                         multicast = stats->multicast + stats->broadcast;
1225                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1226
1227                 t->rx_bytes     += bytes;
1228                 t->rx_packets   += packets;
1229                 t->multicast    += multicast;
1230         }
1231 }
1232
1233 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
1234 {
1235         struct net_device_context *ndc = netdev_priv(ndev);
1236         struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1237         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1238         struct sockaddr *addr = p;
1239         int err;
1240
1241         err = eth_prepare_mac_addr_change(ndev, p);
1242         if (err)
1243                 return err;
1244
1245         if (!nvdev)
1246                 return -ENODEV;
1247
1248         if (vf_netdev) {
1249                 err = dev_set_mac_address(vf_netdev, addr);
1250                 if (err)
1251                         return err;
1252         }
1253
1254         err = rndis_filter_set_device_mac(nvdev, addr->sa_data);
1255         if (!err) {
1256                 eth_commit_mac_addr_change(ndev, p);
1257         } else if (vf_netdev) {
1258                 /* rollback change on VF */
1259                 memcpy(addr->sa_data, ndev->dev_addr, ETH_ALEN);
1260                 dev_set_mac_address(vf_netdev, addr);
1261         }
1262
1263         return err;
1264 }
1265
1266 static const struct {
1267         char name[ETH_GSTRING_LEN];
1268         u16 offset;
1269 } netvsc_stats[] = {
1270         { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
1271         { "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
1272         { "tx_no_space",  offsetof(struct netvsc_ethtool_stats, tx_no_space) },
1273         { "tx_too_big",   offsetof(struct netvsc_ethtool_stats, tx_too_big) },
1274         { "tx_busy",      offsetof(struct netvsc_ethtool_stats, tx_busy) },
1275         { "tx_send_full", offsetof(struct netvsc_ethtool_stats, tx_send_full) },
1276         { "rx_comp_busy", offsetof(struct netvsc_ethtool_stats, rx_comp_busy) },
1277         { "rx_no_memory", offsetof(struct netvsc_ethtool_stats, rx_no_memory) },
1278         { "stop_queue", offsetof(struct netvsc_ethtool_stats, stop_queue) },
1279         { "wake_queue", offsetof(struct netvsc_ethtool_stats, wake_queue) },
1280 }, pcpu_stats[] = {
1281         { "cpu%u_rx_packets",
1282                 offsetof(struct netvsc_ethtool_pcpu_stats, rx_packets) },
1283         { "cpu%u_rx_bytes",
1284                 offsetof(struct netvsc_ethtool_pcpu_stats, rx_bytes) },
1285         { "cpu%u_tx_packets",
1286                 offsetof(struct netvsc_ethtool_pcpu_stats, tx_packets) },
1287         { "cpu%u_tx_bytes",
1288                 offsetof(struct netvsc_ethtool_pcpu_stats, tx_bytes) },
1289         { "cpu%u_vf_rx_packets",
1290                 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_packets) },
1291         { "cpu%u_vf_rx_bytes",
1292                 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_bytes) },
1293         { "cpu%u_vf_tx_packets",
1294                 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_packets) },
1295         { "cpu%u_vf_tx_bytes",
1296                 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_bytes) },
1297 }, vf_stats[] = {
1298         { "vf_rx_packets", offsetof(struct netvsc_vf_pcpu_stats, rx_packets) },
1299         { "vf_rx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, rx_bytes) },
1300         { "vf_tx_packets", offsetof(struct netvsc_vf_pcpu_stats, tx_packets) },
1301         { "vf_tx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, tx_bytes) },
1302         { "vf_tx_dropped", offsetof(struct netvsc_vf_pcpu_stats, tx_dropped) },
1303 };
1304
1305 #define NETVSC_GLOBAL_STATS_LEN ARRAY_SIZE(netvsc_stats)
1306 #define NETVSC_VF_STATS_LEN     ARRAY_SIZE(vf_stats)
1307
1308 /* statistics per queue (rx/tx packets/bytes) */
1309 #define NETVSC_PCPU_STATS_LEN (num_present_cpus() * ARRAY_SIZE(pcpu_stats))
1310
1311 /* 4 statistics per queue (rx/tx packets/bytes) */
1312 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 4)
1313
1314 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
1315 {
1316         struct net_device_context *ndc = netdev_priv(dev);
1317         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1318
1319         if (!nvdev)
1320                 return -ENODEV;
1321
1322         switch (string_set) {
1323         case ETH_SS_STATS:
1324                 return NETVSC_GLOBAL_STATS_LEN
1325                         + NETVSC_VF_STATS_LEN
1326                         + NETVSC_QUEUE_STATS_LEN(nvdev)
1327                         + NETVSC_PCPU_STATS_LEN;
1328         default:
1329                 return -EINVAL;
1330         }
1331 }
1332
1333 static void netvsc_get_ethtool_stats(struct net_device *dev,
1334                                      struct ethtool_stats *stats, u64 *data)
1335 {
1336         struct net_device_context *ndc = netdev_priv(dev);
1337         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1338         const void *nds = &ndc->eth_stats;
1339         const struct netvsc_stats *qstats;
1340         struct netvsc_vf_pcpu_stats sum;
1341         struct netvsc_ethtool_pcpu_stats *pcpu_sum;
1342         unsigned int start;
1343         u64 packets, bytes;
1344         int i, j, cpu;
1345
1346         if (!nvdev)
1347                 return;
1348
1349         for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
1350                 data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1351
1352         netvsc_get_vf_stats(dev, &sum);
1353         for (j = 0; j < NETVSC_VF_STATS_LEN; j++)
1354                 data[i++] = *(u64 *)((void *)&sum + vf_stats[j].offset);
1355
1356         for (j = 0; j < nvdev->num_chn; j++) {
1357                 qstats = &nvdev->chan_table[j].tx_stats;
1358
1359                 do {
1360                         start = u64_stats_fetch_begin_irq(&qstats->syncp);
1361                         packets = qstats->packets;
1362                         bytes = qstats->bytes;
1363                 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1364                 data[i++] = packets;
1365                 data[i++] = bytes;
1366
1367                 qstats = &nvdev->chan_table[j].rx_stats;
1368                 do {
1369                         start = u64_stats_fetch_begin_irq(&qstats->syncp);
1370                         packets = qstats->packets;
1371                         bytes = qstats->bytes;
1372                 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1373                 data[i++] = packets;
1374                 data[i++] = bytes;
1375         }
1376
1377         pcpu_sum = kvmalloc_array(num_possible_cpus(),
1378                                   sizeof(struct netvsc_ethtool_pcpu_stats),
1379                                   GFP_KERNEL);
1380         netvsc_get_pcpu_stats(dev, pcpu_sum);
1381         for_each_present_cpu(cpu) {
1382                 struct netvsc_ethtool_pcpu_stats *this_sum = &pcpu_sum[cpu];
1383
1384                 for (j = 0; j < ARRAY_SIZE(pcpu_stats); j++)
1385                         data[i++] = *(u64 *)((void *)this_sum
1386                                              + pcpu_stats[j].offset);
1387         }
1388         kvfree(pcpu_sum);
1389 }
1390
1391 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1392 {
1393         struct net_device_context *ndc = netdev_priv(dev);
1394         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1395         u8 *p = data;
1396         int i, cpu;
1397
1398         if (!nvdev)
1399                 return;
1400
1401         switch (stringset) {
1402         case ETH_SS_STATS:
1403                 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++) {
1404                         memcpy(p, netvsc_stats[i].name, ETH_GSTRING_LEN);
1405                         p += ETH_GSTRING_LEN;
1406                 }
1407
1408                 for (i = 0; i < ARRAY_SIZE(vf_stats); i++) {
1409                         memcpy(p, vf_stats[i].name, ETH_GSTRING_LEN);
1410                         p += ETH_GSTRING_LEN;
1411                 }
1412
1413                 for (i = 0; i < nvdev->num_chn; i++) {
1414                         sprintf(p, "tx_queue_%u_packets", i);
1415                         p += ETH_GSTRING_LEN;
1416                         sprintf(p, "tx_queue_%u_bytes", i);
1417                         p += ETH_GSTRING_LEN;
1418                         sprintf(p, "rx_queue_%u_packets", i);
1419                         p += ETH_GSTRING_LEN;
1420                         sprintf(p, "rx_queue_%u_bytes", i);
1421                         p += ETH_GSTRING_LEN;
1422                 }
1423
1424                 for_each_present_cpu(cpu) {
1425                         for (i = 0; i < ARRAY_SIZE(pcpu_stats); i++) {
1426                                 sprintf(p, pcpu_stats[i].name, cpu);
1427                                 p += ETH_GSTRING_LEN;
1428                         }
1429                 }
1430
1431                 break;
1432         }
1433 }
1434
1435 static int
1436 netvsc_get_rss_hash_opts(struct net_device_context *ndc,
1437                          struct ethtool_rxnfc *info)
1438 {
1439         const u32 l4_flag = RXH_L4_B_0_1 | RXH_L4_B_2_3;
1440
1441         info->data = RXH_IP_SRC | RXH_IP_DST;
1442
1443         switch (info->flow_type) {
1444         case TCP_V4_FLOW:
1445                 if (ndc->l4_hash & HV_TCP4_L4HASH)
1446                         info->data |= l4_flag;
1447
1448                 break;
1449
1450         case TCP_V6_FLOW:
1451                 if (ndc->l4_hash & HV_TCP6_L4HASH)
1452                         info->data |= l4_flag;
1453
1454                 break;
1455
1456         case UDP_V4_FLOW:
1457                 if (ndc->l4_hash & HV_UDP4_L4HASH)
1458                         info->data |= l4_flag;
1459
1460                 break;
1461
1462         case UDP_V6_FLOW:
1463                 if (ndc->l4_hash & HV_UDP6_L4HASH)
1464                         info->data |= l4_flag;
1465
1466                 break;
1467
1468         case IPV4_FLOW:
1469         case IPV6_FLOW:
1470                 break;
1471         default:
1472                 info->data = 0;
1473                 break;
1474         }
1475
1476         return 0;
1477 }
1478
1479 static int
1480 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1481                  u32 *rules)
1482 {
1483         struct net_device_context *ndc = netdev_priv(dev);
1484         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1485
1486         if (!nvdev)
1487                 return -ENODEV;
1488
1489         switch (info->cmd) {
1490         case ETHTOOL_GRXRINGS:
1491                 info->data = nvdev->num_chn;
1492                 return 0;
1493
1494         case ETHTOOL_GRXFH:
1495                 return netvsc_get_rss_hash_opts(ndc, info);
1496         }
1497         return -EOPNOTSUPP;
1498 }
1499
1500 static int netvsc_set_rss_hash_opts(struct net_device_context *ndc,
1501                                     struct ethtool_rxnfc *info)
1502 {
1503         if (info->data == (RXH_IP_SRC | RXH_IP_DST |
1504                            RXH_L4_B_0_1 | RXH_L4_B_2_3)) {
1505                 switch (info->flow_type) {
1506                 case TCP_V4_FLOW:
1507                         ndc->l4_hash |= HV_TCP4_L4HASH;
1508                         break;
1509
1510                 case TCP_V6_FLOW:
1511                         ndc->l4_hash |= HV_TCP6_L4HASH;
1512                         break;
1513
1514                 case UDP_V4_FLOW:
1515                         ndc->l4_hash |= HV_UDP4_L4HASH;
1516                         break;
1517
1518                 case UDP_V6_FLOW:
1519                         ndc->l4_hash |= HV_UDP6_L4HASH;
1520                         break;
1521
1522                 default:
1523                         return -EOPNOTSUPP;
1524                 }
1525
1526                 return 0;
1527         }
1528
1529         if (info->data == (RXH_IP_SRC | RXH_IP_DST)) {
1530                 switch (info->flow_type) {
1531                 case TCP_V4_FLOW:
1532                         ndc->l4_hash &= ~HV_TCP4_L4HASH;
1533                         break;
1534
1535                 case TCP_V6_FLOW:
1536                         ndc->l4_hash &= ~HV_TCP6_L4HASH;
1537                         break;
1538
1539                 case UDP_V4_FLOW:
1540                         ndc->l4_hash &= ~HV_UDP4_L4HASH;
1541                         break;
1542
1543                 case UDP_V6_FLOW:
1544                         ndc->l4_hash &= ~HV_UDP6_L4HASH;
1545                         break;
1546
1547                 default:
1548                         return -EOPNOTSUPP;
1549                 }
1550
1551                 return 0;
1552         }
1553
1554         return -EOPNOTSUPP;
1555 }
1556
1557 static int
1558 netvsc_set_rxnfc(struct net_device *ndev, struct ethtool_rxnfc *info)
1559 {
1560         struct net_device_context *ndc = netdev_priv(ndev);
1561
1562         if (info->cmd == ETHTOOL_SRXFH)
1563                 return netvsc_set_rss_hash_opts(ndc, info);
1564
1565         return -EOPNOTSUPP;
1566 }
1567
1568 static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1569 {
1570         return NETVSC_HASH_KEYLEN;
1571 }
1572
1573 static u32 netvsc_rss_indir_size(struct net_device *dev)
1574 {
1575         return ITAB_NUM;
1576 }
1577
1578 static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key,
1579                            u8 *hfunc)
1580 {
1581         struct net_device_context *ndc = netdev_priv(dev);
1582         struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1583         struct rndis_device *rndis_dev;
1584         int i;
1585
1586         if (!ndev)
1587                 return -ENODEV;
1588
1589         if (hfunc)
1590                 *hfunc = ETH_RSS_HASH_TOP;      /* Toeplitz */
1591
1592         rndis_dev = ndev->extension;
1593         if (indir) {
1594                 for (i = 0; i < ITAB_NUM; i++)
1595                         indir[i] = rndis_dev->rx_table[i];
1596         }
1597
1598         if (key)
1599                 memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1600
1601         return 0;
1602 }
1603
1604 static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
1605                            const u8 *key, const u8 hfunc)
1606 {
1607         struct net_device_context *ndc = netdev_priv(dev);
1608         struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1609         struct rndis_device *rndis_dev;
1610         int i;
1611
1612         if (!ndev)
1613                 return -ENODEV;
1614
1615         if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
1616                 return -EOPNOTSUPP;
1617
1618         rndis_dev = ndev->extension;
1619         if (indir) {
1620                 for (i = 0; i < ITAB_NUM; i++)
1621                         if (indir[i] >= ndev->num_chn)
1622                                 return -EINVAL;
1623
1624                 for (i = 0; i < ITAB_NUM; i++)
1625                         rndis_dev->rx_table[i] = indir[i];
1626         }
1627
1628         if (!key) {
1629                 if (!indir)
1630                         return 0;
1631
1632                 key = rndis_dev->rss_key;
1633         }
1634
1635         return rndis_filter_set_rss_param(rndis_dev, key);
1636 }
1637
1638 /* Hyper-V RNDIS protocol does not have ring in the HW sense.
1639  * It does have pre-allocated receive area which is divided into sections.
1640  */
1641 static void __netvsc_get_ringparam(struct netvsc_device *nvdev,
1642                                    struct ethtool_ringparam *ring)
1643 {
1644         u32 max_buf_size;
1645
1646         ring->rx_pending = nvdev->recv_section_cnt;
1647         ring->tx_pending = nvdev->send_section_cnt;
1648
1649         if (nvdev->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
1650                 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
1651         else
1652                 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
1653
1654         ring->rx_max_pending = max_buf_size / nvdev->recv_section_size;
1655         ring->tx_max_pending = NETVSC_SEND_BUFFER_SIZE
1656                 / nvdev->send_section_size;
1657 }
1658
1659 static void netvsc_get_ringparam(struct net_device *ndev,
1660                                  struct ethtool_ringparam *ring)
1661 {
1662         struct net_device_context *ndevctx = netdev_priv(ndev);
1663         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1664
1665         if (!nvdev)
1666                 return;
1667
1668         __netvsc_get_ringparam(nvdev, ring);
1669 }
1670
1671 static int netvsc_set_ringparam(struct net_device *ndev,
1672                                 struct ethtool_ringparam *ring)
1673 {
1674         struct net_device_context *ndevctx = netdev_priv(ndev);
1675         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1676         struct netvsc_device_info device_info;
1677         struct ethtool_ringparam orig;
1678         u32 new_tx, new_rx;
1679         int ret = 0;
1680
1681         if (!nvdev || nvdev->destroy)
1682                 return -ENODEV;
1683
1684         memset(&orig, 0, sizeof(orig));
1685         __netvsc_get_ringparam(nvdev, &orig);
1686
1687         new_tx = clamp_t(u32, ring->tx_pending,
1688                          NETVSC_MIN_TX_SECTIONS, orig.tx_max_pending);
1689         new_rx = clamp_t(u32, ring->rx_pending,
1690                          NETVSC_MIN_RX_SECTIONS, orig.rx_max_pending);
1691
1692         if (new_tx == orig.tx_pending &&
1693             new_rx == orig.rx_pending)
1694                 return 0;        /* no change */
1695
1696         memset(&device_info, 0, sizeof(device_info));
1697         device_info.num_chn = nvdev->num_chn;
1698         device_info.send_sections = new_tx;
1699         device_info.send_section_size = nvdev->send_section_size;
1700         device_info.recv_sections = new_rx;
1701         device_info.recv_section_size = nvdev->recv_section_size;
1702
1703         ret = netvsc_detach(ndev, nvdev);
1704         if (ret)
1705                 return ret;
1706
1707         ret = netvsc_attach(ndev, &device_info);
1708         if (ret) {
1709                 device_info.send_sections = orig.tx_pending;
1710                 device_info.recv_sections = orig.rx_pending;
1711
1712                 if (netvsc_attach(ndev, &device_info))
1713                         netdev_err(ndev, "restoring ringparam failed");
1714         }
1715
1716         return ret;
1717 }
1718
1719 static int netvsc_set_features(struct net_device *ndev,
1720                                netdev_features_t features)
1721 {
1722         netdev_features_t change = features ^ ndev->features;
1723         struct net_device_context *ndevctx = netdev_priv(ndev);
1724         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1725         struct ndis_offload_params offloads;
1726
1727         if (!nvdev || nvdev->destroy)
1728                 return -ENODEV;
1729
1730         if (!(change & NETIF_F_LRO))
1731                 return 0;
1732
1733         memset(&offloads, 0, sizeof(struct ndis_offload_params));
1734
1735         if (features & NETIF_F_LRO) {
1736                 offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1737                 offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1738         } else {
1739                 offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1740                 offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1741         }
1742
1743         return rndis_filter_set_offload_params(ndev, nvdev, &offloads);
1744 }
1745
1746 static u32 netvsc_get_msglevel(struct net_device *ndev)
1747 {
1748         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1749
1750         return ndev_ctx->msg_enable;
1751 }
1752
1753 static void netvsc_set_msglevel(struct net_device *ndev, u32 val)
1754 {
1755         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1756
1757         ndev_ctx->msg_enable = val;
1758 }
1759
1760 static const struct ethtool_ops ethtool_ops = {
1761         .get_drvinfo    = netvsc_get_drvinfo,
1762         .get_msglevel   = netvsc_get_msglevel,
1763         .set_msglevel   = netvsc_set_msglevel,
1764         .get_link       = ethtool_op_get_link,
1765         .get_ethtool_stats = netvsc_get_ethtool_stats,
1766         .get_sset_count = netvsc_get_sset_count,
1767         .get_strings    = netvsc_get_strings,
1768         .get_channels   = netvsc_get_channels,
1769         .set_channels   = netvsc_set_channels,
1770         .get_ts_info    = ethtool_op_get_ts_info,
1771         .get_rxnfc      = netvsc_get_rxnfc,
1772         .set_rxnfc      = netvsc_set_rxnfc,
1773         .get_rxfh_key_size = netvsc_get_rxfh_key_size,
1774         .get_rxfh_indir_size = netvsc_rss_indir_size,
1775         .get_rxfh       = netvsc_get_rxfh,
1776         .set_rxfh       = netvsc_set_rxfh,
1777         .get_link_ksettings = netvsc_get_link_ksettings,
1778         .set_link_ksettings = netvsc_set_link_ksettings,
1779         .get_ringparam  = netvsc_get_ringparam,
1780         .set_ringparam  = netvsc_set_ringparam,
1781 };
1782
1783 static const struct net_device_ops device_ops = {
1784         .ndo_open =                     netvsc_open,
1785         .ndo_stop =                     netvsc_close,
1786         .ndo_start_xmit =               netvsc_start_xmit,
1787         .ndo_change_rx_flags =          netvsc_change_rx_flags,
1788         .ndo_set_rx_mode =              netvsc_set_rx_mode,
1789         .ndo_set_features =             netvsc_set_features,
1790         .ndo_change_mtu =               netvsc_change_mtu,
1791         .ndo_validate_addr =            eth_validate_addr,
1792         .ndo_set_mac_address =          netvsc_set_mac_addr,
1793         .ndo_select_queue =             netvsc_select_queue,
1794         .ndo_get_stats64 =              netvsc_get_stats64,
1795 };
1796
1797 /*
1798  * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
1799  * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
1800  * present send GARP packet to network peers with netif_notify_peers().
1801  */
1802 static void netvsc_link_change(struct work_struct *w)
1803 {
1804         struct net_device_context *ndev_ctx =
1805                 container_of(w, struct net_device_context, dwork.work);
1806         struct hv_device *device_obj = ndev_ctx->device_ctx;
1807         struct net_device *net = hv_get_drvdata(device_obj);
1808         struct netvsc_device *net_device;
1809         struct rndis_device *rdev;
1810         struct netvsc_reconfig *event = NULL;
1811         bool notify = false, reschedule = false;
1812         unsigned long flags, next_reconfig, delay;
1813
1814         /* if changes are happening, comeback later */
1815         if (!rtnl_trylock()) {
1816                 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1817                 return;
1818         }
1819
1820         net_device = rtnl_dereference(ndev_ctx->nvdev);
1821         if (!net_device)
1822                 goto out_unlock;
1823
1824         rdev = net_device->extension;
1825
1826         next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
1827         if (time_is_after_jiffies(next_reconfig)) {
1828                 /* link_watch only sends one notification with current state
1829                  * per second, avoid doing reconfig more frequently. Handle
1830                  * wrap around.
1831                  */
1832                 delay = next_reconfig - jiffies;
1833                 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
1834                 schedule_delayed_work(&ndev_ctx->dwork, delay);
1835                 goto out_unlock;
1836         }
1837         ndev_ctx->last_reconfig = jiffies;
1838
1839         spin_lock_irqsave(&ndev_ctx->lock, flags);
1840         if (!list_empty(&ndev_ctx->reconfig_events)) {
1841                 event = list_first_entry(&ndev_ctx->reconfig_events,
1842                                          struct netvsc_reconfig, list);
1843                 list_del(&event->list);
1844                 reschedule = !list_empty(&ndev_ctx->reconfig_events);
1845         }
1846         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1847
1848         if (!event)
1849                 goto out_unlock;
1850
1851         switch (event->event) {
1852                 /* Only the following events are possible due to the check in
1853                  * netvsc_linkstatus_callback()
1854                  */
1855         case RNDIS_STATUS_MEDIA_CONNECT:
1856                 if (rdev->link_state) {
1857                         rdev->link_state = false;
1858                         netif_carrier_on(net);
1859                         netif_tx_wake_all_queues(net);
1860                 } else {
1861                         notify = true;
1862                 }
1863                 kfree(event);
1864                 break;
1865         case RNDIS_STATUS_MEDIA_DISCONNECT:
1866                 if (!rdev->link_state) {
1867                         rdev->link_state = true;
1868                         netif_carrier_off(net);
1869                         netif_tx_stop_all_queues(net);
1870                 }
1871                 kfree(event);
1872                 break;
1873         case RNDIS_STATUS_NETWORK_CHANGE:
1874                 /* Only makes sense if carrier is present */
1875                 if (!rdev->link_state) {
1876                         rdev->link_state = true;
1877                         netif_carrier_off(net);
1878                         netif_tx_stop_all_queues(net);
1879                         event->event = RNDIS_STATUS_MEDIA_CONNECT;
1880                         spin_lock_irqsave(&ndev_ctx->lock, flags);
1881                         list_add(&event->list, &ndev_ctx->reconfig_events);
1882                         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1883                         reschedule = true;
1884                 }
1885                 break;
1886         }
1887
1888         rtnl_unlock();
1889
1890         if (notify)
1891                 netdev_notify_peers(net);
1892
1893         /* link_watch only sends one notification with current state per
1894          * second, handle next reconfig event in 2 seconds.
1895          */
1896         if (reschedule)
1897                 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1898
1899         return;
1900
1901 out_unlock:
1902         rtnl_unlock();
1903 }
1904
1905 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
1906 {
1907         struct net_device_context *net_device_ctx;
1908         struct net_device *dev;
1909
1910         dev = netdev_master_upper_dev_get(vf_netdev);
1911         if (!dev || dev->netdev_ops != &device_ops)
1912                 return NULL;    /* not a netvsc device */
1913
1914         net_device_ctx = netdev_priv(dev);
1915         if (!rtnl_dereference(net_device_ctx->nvdev))
1916                 return NULL;    /* device is removed */
1917
1918         return dev;
1919 }
1920
1921 /* Called when VF is injecting data into network stack.
1922  * Change the associated network device from VF to netvsc.
1923  * note: already called with rcu_read_lock
1924  */
1925 static rx_handler_result_t netvsc_vf_handle_frame(struct sk_buff **pskb)
1926 {
1927         struct sk_buff *skb = *pskb;
1928         struct net_device *ndev = rcu_dereference(skb->dev->rx_handler_data);
1929         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1930         struct netvsc_vf_pcpu_stats *pcpu_stats
1931                  = this_cpu_ptr(ndev_ctx->vf_stats);
1932
1933         skb->dev = ndev;
1934
1935         u64_stats_update_begin(&pcpu_stats->syncp);
1936         pcpu_stats->rx_packets++;
1937         pcpu_stats->rx_bytes += skb->len;
1938         u64_stats_update_end(&pcpu_stats->syncp);
1939
1940         return RX_HANDLER_ANOTHER;
1941 }
1942
1943 static int netvsc_vf_join(struct net_device *vf_netdev,
1944                           struct net_device *ndev)
1945 {
1946         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1947         int ret;
1948
1949         ret = netdev_rx_handler_register(vf_netdev,
1950                                          netvsc_vf_handle_frame, ndev);
1951         if (ret != 0) {
1952                 netdev_err(vf_netdev,
1953                            "can not register netvsc VF receive handler (err = %d)\n",
1954                            ret);
1955                 goto rx_handler_failed;
1956         }
1957
1958         ret = netdev_master_upper_dev_link(vf_netdev, ndev,
1959                                            NULL, NULL, NULL);
1960         if (ret != 0) {
1961                 netdev_err(vf_netdev,
1962                            "can not set master device %s (err = %d)\n",
1963                            ndev->name, ret);
1964                 goto upper_link_failed;
1965         }
1966
1967         /* set slave flag before open to prevent IPv6 addrconf */
1968         vf_netdev->flags |= IFF_SLAVE;
1969
1970         schedule_delayed_work(&ndev_ctx->vf_takeover, VF_TAKEOVER_INT);
1971
1972         call_netdevice_notifiers(NETDEV_JOIN, vf_netdev);
1973
1974         netdev_info(vf_netdev, "joined to %s\n", ndev->name);
1975         return 0;
1976
1977 upper_link_failed:
1978         netdev_rx_handler_unregister(vf_netdev);
1979 rx_handler_failed:
1980         return ret;
1981 }
1982
1983 static void __netvsc_vf_setup(struct net_device *ndev,
1984                               struct net_device *vf_netdev)
1985 {
1986         int ret;
1987
1988         /* Align MTU of VF with master */
1989         ret = dev_set_mtu(vf_netdev, ndev->mtu);
1990         if (ret)
1991                 netdev_warn(vf_netdev,
1992                             "unable to change mtu to %u\n", ndev->mtu);
1993
1994         /* set multicast etc flags on VF */
1995         dev_change_flags(vf_netdev, ndev->flags | IFF_SLAVE);
1996
1997         /* sync address list from ndev to VF */
1998         netif_addr_lock_bh(ndev);
1999         dev_uc_sync(vf_netdev, ndev);
2000         dev_mc_sync(vf_netdev, ndev);
2001         netif_addr_unlock_bh(ndev);
2002
2003         if (netif_running(ndev)) {
2004                 ret = dev_open(vf_netdev);
2005                 if (ret)
2006                         netdev_warn(vf_netdev,
2007                                     "unable to open: %d\n", ret);
2008         }
2009 }
2010
2011 /* Setup VF as slave of the synthetic device.
2012  * Runs in workqueue to avoid recursion in netlink callbacks.
2013  */
2014 static void netvsc_vf_setup(struct work_struct *w)
2015 {
2016         struct net_device_context *ndev_ctx
2017                 = container_of(w, struct net_device_context, vf_takeover.work);
2018         struct net_device *ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2019         struct net_device *vf_netdev;
2020
2021         if (!rtnl_trylock()) {
2022                 schedule_delayed_work(&ndev_ctx->vf_takeover, 0);
2023                 return;
2024         }
2025
2026         vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2027         if (vf_netdev)
2028                 __netvsc_vf_setup(ndev, vf_netdev);
2029
2030         rtnl_unlock();
2031 }
2032
2033 /* Find netvsc by VF serial number.
2034  * The PCI hyperv controller records the serial number as the slot kobj name.
2035  */
2036 static struct net_device *get_netvsc_byslot(const struct net_device *vf_netdev)
2037 {
2038         struct device *parent = vf_netdev->dev.parent;
2039         struct net_device_context *ndev_ctx;
2040         struct pci_dev *pdev;
2041         u32 serial;
2042
2043         if (!parent || !dev_is_pci(parent))
2044                 return NULL; /* not a PCI device */
2045
2046         pdev = to_pci_dev(parent);
2047         if (!pdev->slot) {
2048                 netdev_notice(vf_netdev, "no PCI slot information\n");
2049                 return NULL;
2050         }
2051
2052         if (kstrtou32(pci_slot_name(pdev->slot), 10, &serial)) {
2053                 netdev_notice(vf_netdev, "Invalid vf serial:%s\n",
2054                               pci_slot_name(pdev->slot));
2055                 return NULL;
2056         }
2057
2058         list_for_each_entry(ndev_ctx, &netvsc_dev_list, list) {
2059                 if (!ndev_ctx->vf_alloc)
2060                         continue;
2061
2062                 if (ndev_ctx->vf_serial == serial)
2063                         return hv_get_drvdata(ndev_ctx->device_ctx);
2064         }
2065
2066         netdev_notice(vf_netdev,
2067                       "no netdev found for vf serial:%u\n", serial);
2068         return NULL;
2069 }
2070
2071 static int netvsc_register_vf(struct net_device *vf_netdev)
2072 {
2073         struct net_device_context *net_device_ctx;
2074         struct netvsc_device *netvsc_dev;
2075         struct net_device *ndev;
2076         int ret;
2077
2078         if (vf_netdev->addr_len != ETH_ALEN)
2079                 return NOTIFY_DONE;
2080
2081         ndev = get_netvsc_byslot(vf_netdev);
2082         if (!ndev)
2083                 return NOTIFY_DONE;
2084
2085         net_device_ctx = netdev_priv(ndev);
2086         netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2087         if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
2088                 return NOTIFY_DONE;
2089
2090         /* if syntihetic interface is a different namespace,
2091          * then move the VF to that namespace; join will be
2092          * done again in that context.
2093          */
2094         if (!net_eq(dev_net(ndev), dev_net(vf_netdev))) {
2095                 ret = dev_change_net_namespace(vf_netdev,
2096                                                dev_net(ndev), "eth%d");
2097                 if (ret)
2098                         netdev_err(vf_netdev,
2099                                    "could not move to same namespace as %s: %d\n",
2100                                    ndev->name, ret);
2101                 else
2102                         netdev_info(vf_netdev,
2103                                     "VF moved to namespace with: %s\n",
2104                                     ndev->name);
2105                 return NOTIFY_DONE;
2106         }
2107
2108         netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
2109
2110         if (netvsc_vf_join(vf_netdev, ndev) != 0)
2111                 return NOTIFY_DONE;
2112
2113         dev_hold(vf_netdev);
2114         rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
2115         return NOTIFY_OK;
2116 }
2117
2118 /* VF up/down change detected, schedule to change data path */
2119 static int netvsc_vf_changed(struct net_device *vf_netdev)
2120 {
2121         struct net_device_context *net_device_ctx;
2122         struct netvsc_device *netvsc_dev;
2123         struct net_device *ndev;
2124         bool vf_is_up = netif_running(vf_netdev);
2125
2126         ndev = get_netvsc_byref(vf_netdev);
2127         if (!ndev)
2128                 return NOTIFY_DONE;
2129
2130         net_device_ctx = netdev_priv(ndev);
2131         netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2132         if (!netvsc_dev)
2133                 return NOTIFY_DONE;
2134
2135         netvsc_switch_datapath(ndev, vf_is_up);
2136         netdev_info(ndev, "Data path switched %s VF: %s\n",
2137                     vf_is_up ? "to" : "from", vf_netdev->name);
2138
2139         return NOTIFY_OK;
2140 }
2141
2142 static int netvsc_unregister_vf(struct net_device *vf_netdev)
2143 {
2144         struct net_device *ndev;
2145         struct net_device_context *net_device_ctx;
2146
2147         ndev = get_netvsc_byref(vf_netdev);
2148         if (!ndev)
2149                 return NOTIFY_DONE;
2150
2151         net_device_ctx = netdev_priv(ndev);
2152         cancel_delayed_work_sync(&net_device_ctx->vf_takeover);
2153
2154         netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
2155
2156         netdev_rx_handler_unregister(vf_netdev);
2157         netdev_upper_dev_unlink(vf_netdev, ndev);
2158         RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
2159         dev_put(vf_netdev);
2160
2161         return NOTIFY_OK;
2162 }
2163
2164 static int netvsc_probe(struct hv_device *dev,
2165                         const struct hv_vmbus_device_id *dev_id)
2166 {
2167         struct net_device *net = NULL;
2168         struct net_device_context *net_device_ctx;
2169         struct netvsc_device_info device_info;
2170         struct netvsc_device *nvdev;
2171         int ret = -ENOMEM;
2172
2173         net = alloc_etherdev_mq(sizeof(struct net_device_context),
2174                                 VRSS_CHANNEL_MAX);
2175         if (!net)
2176                 goto no_net;
2177
2178         netif_carrier_off(net);
2179
2180         netvsc_init_settings(net);
2181
2182         net_device_ctx = netdev_priv(net);
2183         net_device_ctx->device_ctx = dev;
2184         net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
2185         if (netif_msg_probe(net_device_ctx))
2186                 netdev_dbg(net, "netvsc msg_enable: %d\n",
2187                            net_device_ctx->msg_enable);
2188
2189         hv_set_drvdata(dev, net);
2190
2191         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
2192
2193         spin_lock_init(&net_device_ctx->lock);
2194         INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
2195         INIT_DELAYED_WORK(&net_device_ctx->vf_takeover, netvsc_vf_setup);
2196
2197         net_device_ctx->vf_stats
2198                 = netdev_alloc_pcpu_stats(struct netvsc_vf_pcpu_stats);
2199         if (!net_device_ctx->vf_stats)
2200                 goto no_stats;
2201
2202         net->netdev_ops = &device_ops;
2203         net->ethtool_ops = &ethtool_ops;
2204         SET_NETDEV_DEV(net, &dev->device);
2205
2206         /* We always need headroom for rndis header */
2207         net->needed_headroom = RNDIS_AND_PPI_SIZE;
2208
2209         /* Initialize the number of queues to be 1, we may change it if more
2210          * channels are offered later.
2211          */
2212         netif_set_real_num_tx_queues(net, 1);
2213         netif_set_real_num_rx_queues(net, 1);
2214
2215         /* Notify the netvsc driver of the new device */
2216         memset(&device_info, 0, sizeof(device_info));
2217         device_info.num_chn = VRSS_CHANNEL_DEFAULT;
2218         device_info.send_sections = NETVSC_DEFAULT_TX;
2219         device_info.send_section_size = NETVSC_SEND_SECTION_SIZE;
2220         device_info.recv_sections = NETVSC_DEFAULT_RX;
2221         device_info.recv_section_size = NETVSC_RECV_SECTION_SIZE;
2222
2223         nvdev = rndis_filter_device_add(dev, &device_info);
2224         if (IS_ERR(nvdev)) {
2225                 ret = PTR_ERR(nvdev);
2226                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
2227                 goto rndis_failed;
2228         }
2229
2230         memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
2231
2232         /* We must get rtnl lock before scheduling nvdev->subchan_work,
2233          * otherwise netvsc_subchan_work() can get rtnl lock first and wait
2234          * all subchannels to show up, but that may not happen because
2235          * netvsc_probe() can't get rtnl lock and as a result vmbus_onoffer()
2236          * -> ... -> device_add() -> ... -> __device_attach() can't get
2237          * the device lock, so all the subchannels can't be processed --
2238          * finally netvsc_subchan_work() hangs for ever.
2239          */
2240         rtnl_lock();
2241
2242         if (nvdev->num_chn > 1)
2243                 schedule_work(&nvdev->subchan_work);
2244
2245         /* hw_features computed in rndis_netdev_set_hwcaps() */
2246         net->features = net->hw_features |
2247                 NETIF_F_HIGHDMA | NETIF_F_SG |
2248                 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
2249         net->vlan_features = net->features;
2250
2251         netdev_lockdep_set_classes(net);
2252
2253         /* MTU range: 68 - 1500 or 65521 */
2254         net->min_mtu = NETVSC_MTU_MIN;
2255         if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
2256                 net->max_mtu = NETVSC_MTU - ETH_HLEN;
2257         else
2258                 net->max_mtu = ETH_DATA_LEN;
2259
2260         ret = register_netdevice(net);
2261         if (ret != 0) {
2262                 pr_err("Unable to register netdev.\n");
2263                 goto register_failed;
2264         }
2265
2266         list_add(&net_device_ctx->list, &netvsc_dev_list);
2267         rtnl_unlock();
2268         return 0;
2269
2270 register_failed:
2271         rtnl_unlock();
2272         rndis_filter_device_remove(dev, nvdev);
2273 rndis_failed:
2274         free_percpu(net_device_ctx->vf_stats);
2275 no_stats:
2276         hv_set_drvdata(dev, NULL);
2277         free_netdev(net);
2278 no_net:
2279         return ret;
2280 }
2281
2282 static int netvsc_remove(struct hv_device *dev)
2283 {
2284         struct net_device_context *ndev_ctx;
2285         struct net_device *vf_netdev, *net;
2286         struct netvsc_device *nvdev;
2287
2288         net = hv_get_drvdata(dev);
2289         if (net == NULL) {
2290                 dev_err(&dev->device, "No net device to remove\n");
2291                 return 0;
2292         }
2293
2294         ndev_ctx = netdev_priv(net);
2295
2296         cancel_delayed_work_sync(&ndev_ctx->dwork);
2297
2298         rtnl_lock();
2299         nvdev = rtnl_dereference(ndev_ctx->nvdev);
2300         if (nvdev)
2301                 cancel_work_sync(&nvdev->subchan_work);
2302
2303         /*
2304          * Call to the vsc driver to let it know that the device is being
2305          * removed. Also blocks mtu and channel changes.
2306          */
2307         vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2308         if (vf_netdev)
2309                 netvsc_unregister_vf(vf_netdev);
2310
2311         if (nvdev)
2312                 rndis_filter_device_remove(dev, nvdev);
2313
2314         unregister_netdevice(net);
2315         list_del(&ndev_ctx->list);
2316
2317         rtnl_unlock();
2318
2319         hv_set_drvdata(dev, NULL);
2320
2321         free_percpu(ndev_ctx->vf_stats);
2322         free_netdev(net);
2323         return 0;
2324 }
2325
2326 static const struct hv_vmbus_device_id id_table[] = {
2327         /* Network guid */
2328         { HV_NIC_GUID, },
2329         { },
2330 };
2331
2332 MODULE_DEVICE_TABLE(vmbus, id_table);
2333
2334 /* The one and only one */
2335 static struct  hv_driver netvsc_drv = {
2336         .name = KBUILD_MODNAME,
2337         .id_table = id_table,
2338         .probe = netvsc_probe,
2339         .remove = netvsc_remove,
2340         .driver = {
2341                 .probe_type = PROBE_PREFER_ASYNCHRONOUS,
2342         },
2343 };
2344
2345 /*
2346  * On Hyper-V, every VF interface is matched with a corresponding
2347  * synthetic interface. The synthetic interface is presented first
2348  * to the guest. When the corresponding VF instance is registered,
2349  * we will take care of switching the data path.
2350  */
2351 static int netvsc_netdev_event(struct notifier_block *this,
2352                                unsigned long event, void *ptr)
2353 {
2354         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2355
2356         /* Skip our own events */
2357         if (event_dev->netdev_ops == &device_ops)
2358                 return NOTIFY_DONE;
2359
2360         /* Avoid non-Ethernet type devices */
2361         if (event_dev->type != ARPHRD_ETHER)
2362                 return NOTIFY_DONE;
2363
2364         /* Avoid Vlan dev with same MAC registering as VF */
2365         if (is_vlan_dev(event_dev))
2366                 return NOTIFY_DONE;
2367
2368         /* Avoid Bonding master dev with same MAC registering as VF */
2369         if ((event_dev->priv_flags & IFF_BONDING) &&
2370             (event_dev->flags & IFF_MASTER))
2371                 return NOTIFY_DONE;
2372
2373         switch (event) {
2374         case NETDEV_REGISTER:
2375                 return netvsc_register_vf(event_dev);
2376         case NETDEV_UNREGISTER:
2377                 return netvsc_unregister_vf(event_dev);
2378         case NETDEV_UP:
2379         case NETDEV_DOWN:
2380                 return netvsc_vf_changed(event_dev);
2381         default:
2382                 return NOTIFY_DONE;
2383         }
2384 }
2385
2386 static struct notifier_block netvsc_netdev_notifier = {
2387         .notifier_call = netvsc_netdev_event,
2388 };
2389
2390 static void __exit netvsc_drv_exit(void)
2391 {
2392         unregister_netdevice_notifier(&netvsc_netdev_notifier);
2393         vmbus_driver_unregister(&netvsc_drv);
2394 }
2395
2396 static int __init netvsc_drv_init(void)
2397 {
2398         int ret;
2399
2400         if (ring_size < RING_SIZE_MIN) {
2401                 ring_size = RING_SIZE_MIN;
2402                 pr_info("Increased ring_size to %u (min allowed)\n",
2403                         ring_size);
2404         }
2405         netvsc_ring_bytes = ring_size * PAGE_SIZE;
2406
2407         ret = vmbus_driver_register(&netvsc_drv);
2408         if (ret)
2409                 return ret;
2410
2411         register_netdevice_notifier(&netvsc_netdev_notifier);
2412         return 0;
2413 }
2414
2415 MODULE_LICENSE("GPL");
2416 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
2417
2418 module_init(netvsc_drv_init);
2419 module_exit(netvsc_drv_exit);