drivers/net/hyperv/netvsc_drv.c

   1 /*
   2  * Copyright (c) 2009, Microsoft Corporation.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms and conditions of the GNU General Public License,
   6  * version 2, as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope it will be useful, but WITHOUT
   9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  11  * more details.
  12  *
  13  * You should have received a copy of the GNU General Public License along with
  14  * this program; if not, see <http://www.gnu.org/licenses/>.
  15  *
  16  * Authors:
  17  *   Haiyang Zhang <haiyangz@microsoft.com>
  18  *   Hank Janssen  <hjanssen@microsoft.com>
  19  */
  20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  21
  22 #include <linux/init.h>
  23 #include <linux/atomic.h>
  24 #include <linux/module.h>
  25 #include <linux/highmem.h>
  26 #include <linux/device.h>
  27 #include <linux/io.h>
  28 #include <linux/delay.h>
  29 #include <linux/netdevice.h>
  30 #include <linux/inetdevice.h>
  31 #include <linux/etherdevice.h>
  32 #include <linux/skbuff.h>
  33 #include <linux/if_vlan.h>
  34 #include <linux/in.h>
  35 #include <linux/slab.h>
  36 #include <net/arp.h>
  37 #include <net/route.h>
  38 #include <net/sock.h>
  39 #include <net/pkt_sched.h>
  40 #include <net/checksum.h>
  41 #include <net/ip6_checksum.h>
  42
  43 #include "hyperv_net.h"
  44
  45 #define RING_SIZE_MIN 64
  46 #define LINKCHANGE_INT (2 * HZ)
  47
  48 static int ring_size = 128;
  49 module_param(ring_size, int, S_IRUGO);
  50 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
  51
  52 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
  53                                 NETIF_MSG_LINK | NETIF_MSG_IFUP |
  54                                 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
  55                                 NETIF_MSG_TX_ERR;
  56
  57 static int debug = -1;
  58 module_param(debug, int, S_IRUGO);
  59 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  60
  61 static void netvsc_set_multicast_list(struct net_device *net)
  62 {
  63         struct net_device_context *net_device_ctx = netdev_priv(net);
  64         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
  65
  66         rndis_filter_update(nvdev);
  67 }
  68
  69 static int netvsc_open(struct net_device *net)
  70 {
  71         struct net_device_context *ndev_ctx = netdev_priv(net);
  72         struct netvsc_device *nvdev = ndev_ctx->nvdev;
  73         struct rndis_device *rdev;
  74         int ret = 0;
  75
  76         netif_carrier_off(net);
  77
  78         /* Open up the device */
  79         ret = rndis_filter_open(nvdev);
  80         if (ret != 0) {
  81                 netdev_err(net, "unable to open device (ret %d).\n", ret);
  82                 return ret;
  83         }
  84
  85         netif_tx_wake_all_queues(net);
  86
  87         rdev = nvdev->extension;
  88         if (!rdev->link_state && !ndev_ctx->datapath)
  89                 netif_carrier_on(net);
  90
  91         return ret;
  92 }
  93
  94 static int netvsc_close(struct net_device *net)
  95 {
  96         struct net_device_context *net_device_ctx = netdev_priv(net);
  97         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
  98         int ret;
  99         u32 aread, i, msec = 10, retry = 0, retry_max = 20;
 100         struct vmbus_channel *chn;
 101
 102         netif_tx_disable(net);
 103
 104         ret = rndis_filter_close(nvdev);
 105         if (ret != 0) {
 106                 netdev_err(net, "unable to close device (ret %d).\n", ret);
 107                 return ret;
 108         }
 109
 110         /* Ensure pending bytes in ring are read */
 111         while (true) {
 112                 aread = 0;
 113                 for (i = 0; i < nvdev->num_chn; i++) {
 114                         chn = nvdev->chan_table[i].channel;
 115                         if (!chn)
 116                                 continue;
 117
 118                         aread = hv_get_bytes_to_read(&chn->inbound);
 119                         if (aread)
 120                                 break;
 121
 122                         aread = hv_get_bytes_to_read(&chn->outbound);
 123                         if (aread)
 124                                 break;
 125                 }
 126
 127                 retry++;
 128                 if (retry > retry_max || aread == 0)
 129                         break;
 130
 131                 msleep(msec);
 132
 133                 if (msec < 1000)
 134                         msec *= 2;
 135         }
 136
 137         if (aread) {
 138                 netdev_err(net, "Ring buffer not empty after closing rndis\n");
 139                 ret = -ETIMEDOUT;
 140         }
 141
 142         return ret;
 143 }
 144
 145 static void *init_ppi_data(struct rndis_message *msg, u32 ppi_size,
 146                                 int pkt_type)
 147 {
 148         struct rndis_packet *rndis_pkt;
 149         struct rndis_per_packet_info *ppi;
 150
 151         rndis_pkt = &msg->msg.pkt;
 152         rndis_pkt->data_offset += ppi_size;
 153
 154         ppi = (struct rndis_per_packet_info *)((void *)rndis_pkt +
 155                 rndis_pkt->per_pkt_info_offset + rndis_pkt->per_pkt_info_len);
 156
 157         ppi->size = ppi_size;
 158         ppi->type = pkt_type;
 159         ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
 160
 161         rndis_pkt->per_pkt_info_len += ppi_size;
 162
 163         return ppi;
 164 }
 165
 166 /* Azure hosts don't support non-TCP port numbers in hashing yet. We compute
 167  * hash for non-TCP traffic with only IP numbers.
 168  */
 169 static inline u32 netvsc_get_hash(struct sk_buff *skb, struct sock *sk)
 170 {
 171         struct flow_keys flow;
 172         u32 hash;
 173         static u32 hashrnd __read_mostly;
 174
 175         net_get_random_once(&hashrnd, sizeof(hashrnd));
 176
 177         if (!skb_flow_dissect_flow_keys(skb, &flow, 0))
 178                 return 0;
 179
 180         if (flow.basic.ip_proto == IPPROTO_TCP) {
 181                 return skb_get_hash(skb);
 182         } else {
 183                 if (flow.basic.n_proto == htons(ETH_P_IP))
 184                         hash = jhash2((u32 *)&flow.addrs.v4addrs, 2, hashrnd);
 185                 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
 186                         hash = jhash2((u32 *)&flow.addrs.v6addrs, 8, hashrnd);
 187                 else
 188                         hash = 0;
 189
 190                 skb_set_hash(skb, hash, PKT_HASH_TYPE_L3);
 191         }
 192
 193         return hash;
 194 }
 195
 196 static inline int netvsc_get_tx_queue(struct net_device *ndev,
 197                                       struct sk_buff *skb, int old_idx)
 198 {
 199         const struct net_device_context *ndc = netdev_priv(ndev);
 200         struct sock *sk = skb->sk;
 201         int q_idx;
 202
 203         q_idx = ndc->tx_send_table[netvsc_get_hash(skb, sk) &
 204                                    (VRSS_SEND_TAB_SIZE - 1)];
 205
 206         /* If queue index changed record the new value */
 207         if (q_idx != old_idx &&
 208             sk && sk_fullsock(sk) && rcu_access_pointer(sk->sk_dst_cache))
 209                 sk_tx_queue_set(sk, q_idx);
 210
 211         return q_idx;
 212 }
 213
 214 /*
 215  * Select queue for transmit.
 216  *
 217  * If a valid queue has already been assigned, then use that.
 218  * Otherwise compute tx queue based on hash and the send table.
 219  *
 220  * This is basically similar to default (__netdev_pick_tx) with the added step
 221  * of using the host send_table when no other queue has been assigned.
 222  *
 223  * TODO support XPS - but get_xps_queue not exported
 224  */
 225 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
 226                         void *accel_priv, select_queue_fallback_t fallback)
 227 {
 228         unsigned int num_tx_queues = ndev->real_num_tx_queues;
 229         int q_idx = sk_tx_queue_get(skb->sk);
 230
 231         if (q_idx < 0 || skb->ooo_okay) {
 232                 /* If forwarding a packet, we use the recorded queue when
 233                  * available for better cache locality.
 234                  */
 235                 if (skb_rx_queue_recorded(skb))
 236                         q_idx = skb_get_rx_queue(skb);
 237                 else
 238                         q_idx = netvsc_get_tx_queue(ndev, skb, q_idx);
 239         }
 240
 241         while (unlikely(q_idx >= num_tx_queues))
 242                 q_idx -= num_tx_queues;
 243
 244         return q_idx;
 245 }
 246
 247 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
 248                         struct hv_page_buffer *pb)
 249 {
 250         int j = 0;
 251
 252         /* Deal with compund pages by ignoring unused part
 253          * of the page.
 254          */
 255         page += (offset >> PAGE_SHIFT);
 256         offset &= ~PAGE_MASK;
 257
 258         while (len > 0) {
 259                 unsigned long bytes;
 260
 261                 bytes = PAGE_SIZE - offset;
 262                 if (bytes > len)
 263                         bytes = len;
 264                 pb[j].pfn = page_to_pfn(page);
 265                 pb[j].offset = offset;
 266                 pb[j].len = bytes;
 267
 268                 offset += bytes;
 269                 len -= bytes;
 270
 271                 if (offset == PAGE_SIZE && len) {
 272                         page++;
 273                         offset = 0;
 274                         j++;
 275                 }
 276         }
 277
 278         return j + 1;
 279 }
 280
 281 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
 282                            struct hv_netvsc_packet *packet,
 283                            struct hv_page_buffer **page_buf)
 284 {
 285         struct hv_page_buffer *pb = *page_buf;
 286         u32 slots_used = 0;
 287         char *data = skb->data;
 288         int frags = skb_shinfo(skb)->nr_frags;
 289         int i;
 290
 291         /* The packet is laid out thus:
 292          * 1. hdr: RNDIS header and PPI
 293          * 2. skb linear data
 294          * 3. skb fragment data
 295          */
 296         if (hdr != NULL)
 297                 slots_used += fill_pg_buf(virt_to_page(hdr),
 298                                         offset_in_page(hdr),
 299                                         len, &pb[slots_used]);
 300
 301         packet->rmsg_size = len;
 302         packet->rmsg_pgcnt = slots_used;
 303
 304         slots_used += fill_pg_buf(virt_to_page(data),
 305                                 offset_in_page(data),
 306                                 skb_headlen(skb), &pb[slots_used]);
 307
 308         for (i = 0; i < frags; i++) {
 309                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 310
 311                 slots_used += fill_pg_buf(skb_frag_page(frag),
 312                                         frag->page_offset,
 313                                         skb_frag_size(frag), &pb[slots_used]);
 314         }
 315         return slots_used;
 316 }
 317
 318 static int count_skb_frag_slots(struct sk_buff *skb)
 319 {
 320         int i, frags = skb_shinfo(skb)->nr_frags;
 321         int pages = 0;
 322
 323         for (i = 0; i < frags; i++) {
 324                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 325                 unsigned long size = skb_frag_size(frag);
 326                 unsigned long offset = frag->page_offset;
 327
 328                 /* Skip unused frames from start of page */
 329                 offset &= ~PAGE_MASK;
 330                 pages += PFN_UP(offset + size);
 331         }
 332         return pages;
 333 }
 334
 335 static int netvsc_get_slots(struct sk_buff *skb)
 336 {
 337         char *data = skb->data;
 338         unsigned int offset = offset_in_page(data);
 339         unsigned int len = skb_headlen(skb);
 340         int slots;
 341         int frag_slots;
 342
 343         slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
 344         frag_slots = count_skb_frag_slots(skb);
 345         return slots + frag_slots;
 346 }
 347
 348 static u32 net_checksum_info(struct sk_buff *skb)
 349 {
 350         if (skb->protocol == htons(ETH_P_IP)) {
 351                 struct iphdr *ip = ip_hdr(skb);
 352
 353                 if (ip->protocol == IPPROTO_TCP)
 354                         return TRANSPORT_INFO_IPV4_TCP;
 355                 else if (ip->protocol == IPPROTO_UDP)
 356                         return TRANSPORT_INFO_IPV4_UDP;
 357         } else {
 358                 struct ipv6hdr *ip6 = ipv6_hdr(skb);
 359
 360                 if (ip6->nexthdr == IPPROTO_TCP)
 361                         return TRANSPORT_INFO_IPV6_TCP;
 362                 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
 363                         return TRANSPORT_INFO_IPV6_UDP;
 364         }
 365
 366         return TRANSPORT_INFO_NOT_IP;
 367 }
 368
 369 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
 370 {
 371         struct net_device_context *net_device_ctx = netdev_priv(net);
 372         struct hv_netvsc_packet *packet = NULL;
 373         int ret;
 374         unsigned int num_data_pgs;
 375         struct rndis_message *rndis_msg;
 376         struct rndis_packet *rndis_pkt;
 377         u32 rndis_msg_size;
 378         struct rndis_per_packet_info *ppi;
 379         u32 hash;
 380         struct hv_page_buffer page_buf[MAX_PAGE_BUFFER_COUNT];
 381         struct hv_page_buffer *pb = page_buf;
 382
 383         /* We will atmost need two pages to describe the rndis
 384          * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
 385          * of pages in a single packet. If skb is scattered around
 386          * more pages we try linearizing it.
 387          */
 388
 389         num_data_pgs = netvsc_get_slots(skb) + 2;
 390
 391         if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
 392                 ++net_device_ctx->eth_stats.tx_scattered;
 393
 394                 if (skb_linearize(skb))
 395                         goto no_memory;
 396
 397                 num_data_pgs = netvsc_get_slots(skb) + 2;
 398                 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
 399                         ++net_device_ctx->eth_stats.tx_too_big;
 400                         goto drop;
 401                 }
 402         }
 403
 404         /*
 405          * Place the rndis header in the skb head room and
 406          * the skb->cb will be used for hv_netvsc_packet
 407          * structure.
 408          */
 409         ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
 410         if (ret)
 411                 goto no_memory;
 412
 413         /* Use the skb control buffer for building up the packet */
 414         BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
 415                         FIELD_SIZEOF(struct sk_buff, cb));
 416         packet = (struct hv_netvsc_packet *)skb->cb;
 417
 418         packet->q_idx = skb_get_queue_mapping(skb);
 419
 420         packet->total_data_buflen = skb->len;
 421         packet->total_bytes = skb->len;
 422         packet->total_packets = 1;
 423
 424         rndis_msg = (struct rndis_message *)skb->head;
 425
 426         memset(rndis_msg, 0, RNDIS_AND_PPI_SIZE);
 427
 428         /* Add the rndis header */
 429         rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
 430         rndis_msg->msg_len = packet->total_data_buflen;
 431         rndis_pkt = &rndis_msg->msg.pkt;
 432         rndis_pkt->data_offset = sizeof(struct rndis_packet);
 433         rndis_pkt->data_len = packet->total_data_buflen;
 434         rndis_pkt->per_pkt_info_offset = sizeof(struct rndis_packet);
 435
 436         rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
 437
 438         hash = skb_get_hash_raw(skb);
 439         if (hash != 0 && net->real_num_tx_queues > 1) {
 440                 rndis_msg_size += NDIS_HASH_PPI_SIZE;
 441                 ppi = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
 442                                     NBL_HASH_VALUE);
 443                 *(u32 *)((void *)ppi + ppi->ppi_offset) = hash;
 444         }
 445
 446         if (skb_vlan_tag_present(skb)) {
 447                 struct ndis_pkt_8021q_info *vlan;
 448
 449                 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
 450                 ppi = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
 451                                         IEEE_8021Q_INFO);
 452                 vlan = (struct ndis_pkt_8021q_info *)((void *)ppi +
 453                                                 ppi->ppi_offset);
 454                 vlan->vlanid = skb->vlan_tci & VLAN_VID_MASK;
 455                 vlan->pri = (skb->vlan_tci & VLAN_PRIO_MASK) >>
 456                                 VLAN_PRIO_SHIFT;
 457         }
 458
 459         if (skb_is_gso(skb)) {
 460                 struct ndis_tcp_lso_info *lso_info;
 461
 462                 rndis_msg_size += NDIS_LSO_PPI_SIZE;
 463                 ppi = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
 464                                     TCP_LARGESEND_PKTINFO);
 465
 466                 lso_info = (struct ndis_tcp_lso_info *)((void *)ppi +
 467                                                         ppi->ppi_offset);
 468
 469                 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
 470                 if (skb->protocol == htons(ETH_P_IP)) {
 471                         lso_info->lso_v2_transmit.ip_version =
 472                                 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
 473                         ip_hdr(skb)->tot_len = 0;
 474                         ip_hdr(skb)->check = 0;
 475                         tcp_hdr(skb)->check =
 476                                 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
 477                                                    ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 478                 } else {
 479                         lso_info->lso_v2_transmit.ip_version =
 480                                 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
 481                         ipv6_hdr(skb)->payload_len = 0;
 482                         tcp_hdr(skb)->check =
 483                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 484                                                  &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 485                 }
 486                 lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb);
 487                 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
 488         } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
 489                 if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) {
 490                         struct ndis_tcp_ip_checksum_info *csum_info;
 491
 492                         rndis_msg_size += NDIS_CSUM_PPI_SIZE;
 493                         ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
 494                                             TCPIP_CHKSUM_PKTINFO);
 495
 496                         csum_info = (struct ndis_tcp_ip_checksum_info *)((void *)ppi +
 497                                                                          ppi->ppi_offset);
 498
 499                         csum_info->transmit.tcp_header_offset = skb_transport_offset(skb);
 500
 501                         if (skb->protocol == htons(ETH_P_IP)) {
 502                                 csum_info->transmit.is_ipv4 = 1;
 503
 504                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
 505                                         csum_info->transmit.tcp_checksum = 1;
 506                                 else
 507                                         csum_info->transmit.udp_checksum = 1;
 508                         } else {
 509                                 csum_info->transmit.is_ipv6 = 1;
 510
 511                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
 512                                         csum_info->transmit.tcp_checksum = 1;
 513                                 else
 514                                         csum_info->transmit.udp_checksum = 1;
 515                         }
 516                 } else {
 517                         /* Can't do offload of this type of checksum */
 518                         if (skb_checksum_help(skb))
 519                                 goto drop;
 520                 }
 521         }
 522
 523         /* Start filling in the page buffers with the rndis hdr */
 524         rndis_msg->msg_len += rndis_msg_size;
 525         packet->total_data_buflen = rndis_msg->msg_len;
 526         packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
 527                                                skb, packet, &pb);
 528
 529         /* timestamp packet in software */
 530         skb_tx_timestamp(skb);
 531         ret = netvsc_send(net_device_ctx->device_ctx, packet,
 532                           rndis_msg, &pb, skb);
 533         if (likely(ret == 0))
 534                 return NETDEV_TX_OK;
 535
 536         if (ret == -EAGAIN) {
 537                 ++net_device_ctx->eth_stats.tx_busy;
 538                 return NETDEV_TX_BUSY;
 539         }
 540
 541         if (ret == -ENOSPC)
 542                 ++net_device_ctx->eth_stats.tx_no_space;
 543
 544 drop:
 545         dev_kfree_skb_any(skb);
 546         net->stats.tx_dropped++;
 547
 548         return NETDEV_TX_OK;
 549
 550 no_memory:
 551         ++net_device_ctx->eth_stats.tx_no_memory;
 552         goto drop;
 553 }
 554 /*
 555  * netvsc_linkstatus_callback - Link up/down notification
 556  */
 557 void netvsc_linkstatus_callback(struct hv_device *device_obj,
 558                                 struct rndis_message *resp)
 559 {
 560         struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
 561         struct net_device *net;
 562         struct net_device_context *ndev_ctx;
 563         struct netvsc_reconfig *event;
 564         unsigned long flags;
 565
 566         net = hv_get_drvdata(device_obj);
 567
 568         if (!net)
 569                 return;
 570
 571         ndev_ctx = netdev_priv(net);
 572
 573         /* Update the physical link speed when changing to another vSwitch */
 574         if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
 575                 u32 speed;
 576
 577                 speed = *(u32 *)((void *)indicate + indicate->
 578                                  status_buf_offset) / 10000;
 579                 ndev_ctx->speed = speed;
 580                 return;
 581         }
 582
 583         /* Handle these link change statuses below */
 584         if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
 585             indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
 586             indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
 587                 return;
 588
 589         if (net->reg_state != NETREG_REGISTERED)
 590                 return;
 591
 592         event = kzalloc(sizeof(*event), GFP_ATOMIC);
 593         if (!event)
 594                 return;
 595         event->event = indicate->status;
 596
 597         spin_lock_irqsave(&ndev_ctx->lock, flags);
 598         list_add_tail(&event->list, &ndev_ctx->reconfig_events);
 599         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
 600
 601         schedule_delayed_work(&ndev_ctx->dwork, 0);
 602 }
 603
 604 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
 605                                              struct napi_struct *napi,
 606                                              const struct ndis_tcp_ip_checksum_info *csum_info,
 607                                              const struct ndis_pkt_8021q_info *vlan,
 608                                              void *data, u32 buflen)
 609 {
 610         struct sk_buff *skb;
 611
 612         skb = napi_alloc_skb(napi, buflen);
 613         if (!skb)
 614                 return skb;
 615
 616         /*
 617          * Copy to skb. This copy is needed here since the memory pointed by
 618          * hv_netvsc_packet cannot be deallocated
 619          */
 620         skb_put_data(skb, data, buflen);
 621
 622         skb->protocol = eth_type_trans(skb, net);
 623
 624         /* skb is already created with CHECKSUM_NONE */
 625         skb_checksum_none_assert(skb);
 626
 627         /*
 628          * In Linux, the IP checksum is always checked.
 629          * Do L4 checksum offload if enabled and present.
 630          */
 631         if (csum_info && (net->features & NETIF_F_RXCSUM)) {
 632                 if (csum_info->receive.tcp_checksum_succeeded ||
 633                     csum_info->receive.udp_checksum_succeeded)
 634                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 635         }
 636
 637         if (vlan) {
 638                 u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT);
 639
 640                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 641                                        vlan_tci);
 642         }
 643
 644         return skb;
 645 }
 646
 647 /*
 648  * netvsc_recv_callback -  Callback when we receive a packet from the
 649  * "wire" on the specified device.
 650  */
 651 int netvsc_recv_callback(struct net_device *net,
 652                          struct vmbus_channel *channel,
 653                          void  *data, u32 len,
 654                          const struct ndis_tcp_ip_checksum_info *csum_info,
 655                          const struct ndis_pkt_8021q_info *vlan)
 656 {
 657         struct net_device_context *net_device_ctx = netdev_priv(net);
 658         struct netvsc_device *net_device;
 659         u16 q_idx = channel->offermsg.offer.sub_channel_index;
 660         struct netvsc_channel *nvchan;
 661         struct net_device *vf_netdev;
 662         struct sk_buff *skb;
 663         struct netvsc_stats *rx_stats;
 664
 665         if (net->reg_state != NETREG_REGISTERED)
 666                 return NVSP_STAT_FAIL;
 667
 668         /*
 669          * If necessary, inject this packet into the VF interface.
 670          * On Hyper-V, multicast and brodcast packets are only delivered
 671          * to the synthetic interface (after subjecting these to
 672          * policy filters on the host). Deliver these via the VF
 673          * interface in the guest.
 674          */
 675         rcu_read_lock();
 676         net_device = rcu_dereference(net_device_ctx->nvdev);
 677         if (unlikely(!net_device))
 678                 goto drop;
 679
 680         nvchan = &net_device->chan_table[q_idx];
 681         vf_netdev = rcu_dereference(net_device_ctx->vf_netdev);
 682         if (vf_netdev && (vf_netdev->flags & IFF_UP))
 683                 net = vf_netdev;
 684
 685         /* Allocate a skb - TODO direct I/O to pages? */
 686         skb = netvsc_alloc_recv_skb(net, &nvchan->napi,
 687                                     csum_info, vlan, data, len);
 688         if (unlikely(!skb)) {
 689 drop:
 690                 ++net->stats.rx_dropped;
 691                 rcu_read_unlock();
 692                 return NVSP_STAT_FAIL;
 693         }
 694
 695         if (net != vf_netdev)
 696                 skb_record_rx_queue(skb, q_idx);
 697
 698         /*
 699          * Even if injecting the packet, record the statistics
 700          * on the synthetic device because modifying the VF device
 701          * statistics will not work correctly.
 702          */
 703         rx_stats = &nvchan->rx_stats;
 704         u64_stats_update_begin(&rx_stats->syncp);
 705         rx_stats->packets++;
 706         rx_stats->bytes += len;
 707
 708         if (skb->pkt_type == PACKET_BROADCAST)
 709                 ++rx_stats->broadcast;
 710         else if (skb->pkt_type == PACKET_MULTICAST)
 711                 ++rx_stats->multicast;
 712         u64_stats_update_end(&rx_stats->syncp);
 713
 714         napi_gro_receive(&nvchan->napi, skb);
 715         rcu_read_unlock();
 716
 717         return 0;
 718 }
 719
 720 static void netvsc_get_drvinfo(struct net_device *net,
 721                                struct ethtool_drvinfo *info)
 722 {
 723         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
 724         strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
 725 }
 726
 727 static void netvsc_get_channels(struct net_device *net,
 728                                 struct ethtool_channels *channel)
 729 {
 730         struct net_device_context *net_device_ctx = netdev_priv(net);
 731         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
 732
 733         if (nvdev) {
 734                 channel->max_combined   = nvdev->max_chn;
 735                 channel->combined_count = nvdev->num_chn;
 736         }
 737 }
 738
 739 static int netvsc_set_queues(struct net_device *net, struct hv_device *dev,
 740                              u32 num_chn)
 741 {
 742         struct netvsc_device_info device_info;
 743         int ret;
 744
 745         memset(&device_info, 0, sizeof(device_info));
 746         device_info.num_chn = num_chn;
 747         device_info.ring_size = ring_size;
 748         device_info.max_num_vrss_chns = num_chn;
 749
 750         ret = rndis_filter_device_add(dev, &device_info);
 751         if (ret)
 752                 return ret;
 753
 754         ret = netif_set_real_num_tx_queues(net, num_chn);
 755         if (ret)
 756                 return ret;
 757
 758         ret = netif_set_real_num_rx_queues(net, num_chn);
 759
 760         return ret;
 761 }
 762
 763 static int netvsc_set_channels(struct net_device *net,
 764                                struct ethtool_channels *channels)
 765 {
 766         struct net_device_context *net_device_ctx = netdev_priv(net);
 767         struct hv_device *dev = net_device_ctx->device_ctx;
 768         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
 769         unsigned int count = channels->combined_count;
 770         bool was_running;
 771         int ret;
 772
 773         /* We do not support separate count for rx, tx, or other */
 774         if (count == 0 ||
 775             channels->rx_count || channels->tx_count || channels->other_count)
 776                 return -EINVAL;
 777
 778         if (count > net->num_tx_queues || count > VRSS_CHANNEL_MAX)
 779                 return -EINVAL;
 780
 781         if (!nvdev || nvdev->destroy)
 782                 return -ENODEV;
 783
 784         if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
 785                 return -EINVAL;
 786
 787         if (count > nvdev->max_chn)
 788                 return -EINVAL;
 789
 790         was_running = netif_running(net);
 791         if (was_running) {
 792                 ret = netvsc_close(net);
 793                 if (ret)
 794                         return ret;
 795         }
 796
 797         rndis_filter_device_remove(dev, nvdev);
 798
 799         ret = netvsc_set_queues(net, dev, count);
 800         if (ret == 0)
 801                 nvdev->num_chn = count;
 802         else
 803                 netvsc_set_queues(net, dev, nvdev->num_chn);
 804
 805         if (was_running)
 806                 ret = netvsc_open(net);
 807
 808         /* We may have missed link change notifications */
 809         schedule_delayed_work(&net_device_ctx->dwork, 0);
 810
 811         return ret;
 812 }
 813
 814 static bool
 815 netvsc_validate_ethtool_ss_cmd(const struct ethtool_link_ksettings *cmd)
 816 {
 817         struct ethtool_link_ksettings diff1 = *cmd;
 818         struct ethtool_link_ksettings diff2 = {};
 819
 820         diff1.base.speed = 0;
 821         diff1.base.duplex = 0;
 822         /* advertising and cmd are usually set */
 823         ethtool_link_ksettings_zero_link_mode(&diff1, advertising);
 824         diff1.base.cmd = 0;
 825         /* We set port to PORT_OTHER */
 826         diff2.base.port = PORT_OTHER;
 827
 828         return !memcmp(&diff1, &diff2, sizeof(diff1));
 829 }
 830
 831 static void netvsc_init_settings(struct net_device *dev)
 832 {
 833         struct net_device_context *ndc = netdev_priv(dev);
 834
 835         ndc->speed = SPEED_UNKNOWN;
 836         ndc->duplex = DUPLEX_FULL;
 837 }
 838
 839 static int netvsc_get_link_ksettings(struct net_device *dev,
 840                                      struct ethtool_link_ksettings *cmd)
 841 {
 842         struct net_device_context *ndc = netdev_priv(dev);
 843
 844         cmd->base.speed = ndc->speed;
 845         cmd->base.duplex = ndc->duplex;
 846         cmd->base.port = PORT_OTHER;
 847
 848         return 0;
 849 }
 850
 851 static int netvsc_set_link_ksettings(struct net_device *dev,
 852                                      const struct ethtool_link_ksettings *cmd)
 853 {
 854         struct net_device_context *ndc = netdev_priv(dev);
 855         u32 speed;
 856
 857         speed = cmd->base.speed;
 858         if (!ethtool_validate_speed(speed) ||
 859             !ethtool_validate_duplex(cmd->base.duplex) ||
 860             !netvsc_validate_ethtool_ss_cmd(cmd))
 861                 return -EINVAL;
 862
 863         ndc->speed = speed;
 864         ndc->duplex = cmd->base.duplex;
 865
 866         return 0;
 867 }
 868
 869 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
 870 {
 871         struct net_device_context *ndevctx = netdev_priv(ndev);
 872         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
 873         struct hv_device *hdev = ndevctx->device_ctx;
 874         struct netvsc_device_info device_info;
 875         bool was_running;
 876         int ret = 0;
 877
 878         if (!nvdev || nvdev->destroy)
 879                 return -ENODEV;
 880
 881         was_running = netif_running(ndev);
 882         if (was_running) {
 883                 ret = netvsc_close(ndev);
 884                 if (ret)
 885                         return ret;
 886         }
 887
 888         memset(&device_info, 0, sizeof(device_info));
 889         device_info.ring_size = ring_size;
 890         device_info.num_chn = nvdev->num_chn;
 891         device_info.max_num_vrss_chns = nvdev->num_chn;
 892
 893         rndis_filter_device_remove(hdev, nvdev);
 894
 895         /* 'nvdev' has been freed in rndis_filter_device_remove() ->
 896          * netvsc_device_remove () -> free_netvsc_device().
 897          * We mustn't access it before it's re-created in
 898          * rndis_filter_device_add() -> netvsc_device_add().
 899          */
 900
 901         ndev->mtu = mtu;
 902
 903         rndis_filter_device_add(hdev, &device_info);
 904
 905         if (was_running)
 906                 ret = netvsc_open(ndev);
 907
 908         /* We may have missed link change notifications */
 909         schedule_delayed_work(&ndevctx->dwork, 0);
 910
 911         return ret;
 912 }
 913
 914 static void netvsc_get_stats64(struct net_device *net,
 915                                struct rtnl_link_stats64 *t)
 916 {
 917         struct net_device_context *ndev_ctx = netdev_priv(net);
 918         struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
 919         int i;
 920
 921         if (!nvdev)
 922                 return;
 923
 924         for (i = 0; i < nvdev->num_chn; i++) {
 925                 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
 926                 const struct netvsc_stats *stats;
 927                 u64 packets, bytes, multicast;
 928                 unsigned int start;
 929
 930                 stats = &nvchan->tx_stats;
 931                 do {
 932                         start = u64_stats_fetch_begin_irq(&stats->syncp);
 933                         packets = stats->packets;
 934                         bytes = stats->bytes;
 935                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
 936
 937                 t->tx_bytes     += bytes;
 938                 t->tx_packets   += packets;
 939
 940                 stats = &nvchan->rx_stats;
 941                 do {
 942                         start = u64_stats_fetch_begin_irq(&stats->syncp);
 943                         packets = stats->packets;
 944                         bytes = stats->bytes;
 945                         multicast = stats->multicast + stats->broadcast;
 946                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
 947
 948                 t->rx_bytes     += bytes;
 949                 t->rx_packets   += packets;
 950                 t->multicast    += multicast;
 951         }
 952
 953         t->tx_dropped   = net->stats.tx_dropped;
 954         t->tx_errors    = net->stats.tx_errors;
 955
 956         t->rx_dropped   = net->stats.rx_dropped;
 957         t->rx_errors    = net->stats.rx_errors;
 958 }
 959
 960 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
 961 {
 962         struct sockaddr *addr = p;
 963         char save_adr[ETH_ALEN];
 964         unsigned char save_aatype;
 965         int err;
 966
 967         memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
 968         save_aatype = ndev->addr_assign_type;
 969
 970         err = eth_mac_addr(ndev, p);
 971         if (err != 0)
 972                 return err;
 973
 974         err = rndis_filter_set_device_mac(ndev, addr->sa_data);
 975         if (err != 0) {
 976                 /* roll back to saved MAC */
 977                 memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
 978                 ndev->addr_assign_type = save_aatype;
 979         }
 980
 981         return err;
 982 }
 983
 984 static const struct {
 985         char name[ETH_GSTRING_LEN];
 986         u16 offset;
 987 } netvsc_stats[] = {
 988         { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
 989         { "tx_no_memory",  offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
 990         { "tx_no_space",  offsetof(struct netvsc_ethtool_stats, tx_no_space) },
 991         { "tx_too_big",   offsetof(struct netvsc_ethtool_stats, tx_too_big) },
 992         { "tx_busy",      offsetof(struct netvsc_ethtool_stats, tx_busy) },
 993 };
 994
 995 #define NETVSC_GLOBAL_STATS_LEN ARRAY_SIZE(netvsc_stats)
 996
 997 /* 4 statistics per queue (rx/tx packets/bytes) */
 998 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 4)
 999
1000 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
1001 {
1002         struct net_device_context *ndc = netdev_priv(dev);
1003         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1004
1005         if (!nvdev)
1006                 return -ENODEV;
1007
1008         switch (string_set) {
1009         case ETH_SS_STATS:
1010                 return NETVSC_GLOBAL_STATS_LEN + NETVSC_QUEUE_STATS_LEN(nvdev);
1011         default:
1012                 return -EINVAL;
1013         }
1014 }
1015
1016 static void netvsc_get_ethtool_stats(struct net_device *dev,
1017                                      struct ethtool_stats *stats, u64 *data)
1018 {
1019         struct net_device_context *ndc = netdev_priv(dev);
1020         struct netvsc_device *nvdev = rcu_dereference(ndc->nvdev);
1021         const void *nds = &ndc->eth_stats;
1022         const struct netvsc_stats *qstats;
1023         unsigned int start;
1024         u64 packets, bytes;
1025         int i, j;
1026
1027         if (!nvdev)
1028                 return;
1029
1030         for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
1031                 data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1032
1033         for (j = 0; j < nvdev->num_chn; j++) {
1034                 qstats = &nvdev->chan_table[j].tx_stats;
1035
1036                 do {
1037                         start = u64_stats_fetch_begin_irq(&qstats->syncp);
1038                         packets = qstats->packets;
1039                         bytes = qstats->bytes;
1040                 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1041                 data[i++] = packets;
1042                 data[i++] = bytes;
1043
1044                 qstats = &nvdev->chan_table[j].rx_stats;
1045                 do {
1046                         start = u64_stats_fetch_begin_irq(&qstats->syncp);
1047                         packets = qstats->packets;
1048                         bytes = qstats->bytes;
1049                 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1050                 data[i++] = packets;
1051                 data[i++] = bytes;
1052         }
1053 }
1054
1055 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1056 {
1057         struct net_device_context *ndc = netdev_priv(dev);
1058         struct netvsc_device *nvdev = rcu_dereference(ndc->nvdev);
1059         u8 *p = data;
1060         int i;
1061
1062         if (!nvdev)
1063                 return;
1064
1065         switch (stringset) {
1066         case ETH_SS_STATS:
1067                 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++)
1068                         memcpy(p + i * ETH_GSTRING_LEN,
1069                                netvsc_stats[i].name, ETH_GSTRING_LEN);
1070
1071                 p += i * ETH_GSTRING_LEN;
1072                 for (i = 0; i < nvdev->num_chn; i++) {
1073                         sprintf(p, "tx_queue_%u_packets", i);
1074                         p += ETH_GSTRING_LEN;
1075                         sprintf(p, "tx_queue_%u_bytes", i);
1076                         p += ETH_GSTRING_LEN;
1077                         sprintf(p, "rx_queue_%u_packets", i);
1078                         p += ETH_GSTRING_LEN;
1079                         sprintf(p, "rx_queue_%u_bytes", i);
1080                         p += ETH_GSTRING_LEN;
1081                 }
1082
1083                 break;
1084         }
1085 }
1086
1087 static int
1088 netvsc_get_rss_hash_opts(struct netvsc_device *nvdev,
1089                          struct ethtool_rxnfc *info)
1090 {
1091         info->data = RXH_IP_SRC | RXH_IP_DST;
1092
1093         switch (info->flow_type) {
1094         case TCP_V4_FLOW:
1095         case TCP_V6_FLOW:
1096                 info->data |= RXH_L4_B_0_1 | RXH_L4_B_2_3;
1097                 /* fallthrough */
1098         case UDP_V4_FLOW:
1099         case UDP_V6_FLOW:
1100         case IPV4_FLOW:
1101         case IPV6_FLOW:
1102                 break;
1103         default:
1104                 info->data = 0;
1105                 break;
1106         }
1107
1108         return 0;
1109 }
1110
1111 static int
1112 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1113                  u32 *rules)
1114 {
1115         struct net_device_context *ndc = netdev_priv(dev);
1116         struct netvsc_device *nvdev = rcu_dereference(ndc->nvdev);
1117
1118         if (!nvdev)
1119                 return -ENODEV;
1120
1121         switch (info->cmd) {
1122         case ETHTOOL_GRXRINGS:
1123                 info->data = nvdev->num_chn;
1124                 return 0;
1125
1126         case ETHTOOL_GRXFH:
1127                 return netvsc_get_rss_hash_opts(nvdev, info);
1128         }
1129         return -EOPNOTSUPP;
1130 }
1131
1132 #ifdef CONFIG_NET_POLL_CONTROLLER
1133 static void netvsc_poll_controller(struct net_device *dev)
1134 {
1135         struct net_device_context *ndc = netdev_priv(dev);
1136         struct netvsc_device *ndev;
1137         int i;
1138
1139         rcu_read_lock();
1140         ndev = rcu_dereference(ndc->nvdev);
1141         if (ndev) {
1142                 for (i = 0; i < ndev->num_chn; i++) {
1143                         struct netvsc_channel *nvchan = &ndev->chan_table[i];
1144
1145                         napi_schedule(&nvchan->napi);
1146                 }
1147         }
1148         rcu_read_unlock();
1149 }
1150 #endif
1151
1152 static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1153 {
1154         return NETVSC_HASH_KEYLEN;
1155 }
1156
1157 static u32 netvsc_rss_indir_size(struct net_device *dev)
1158 {
1159         return ITAB_NUM;
1160 }
1161
1162 static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key,
1163                            u8 *hfunc)
1164 {
1165         struct net_device_context *ndc = netdev_priv(dev);
1166         struct netvsc_device *ndev = rcu_dereference(ndc->nvdev);
1167         struct rndis_device *rndis_dev;
1168         int i;
1169
1170         if (!ndev)
1171                 return -ENODEV;
1172
1173         if (hfunc)
1174                 *hfunc = ETH_RSS_HASH_TOP;      /* Toeplitz */
1175
1176         rndis_dev = ndev->extension;
1177         if (indir) {
1178                 for (i = 0; i < ITAB_NUM; i++)
1179                         indir[i] = rndis_dev->ind_table[i];
1180         }
1181
1182         if (key)
1183                 memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1184
1185         return 0;
1186 }
1187
1188 static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
1189                            const u8 *key, const u8 hfunc)
1190 {
1191         struct net_device_context *ndc = netdev_priv(dev);
1192         struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1193         struct rndis_device *rndis_dev;
1194         int i;
1195
1196         if (!ndev)
1197                 return -ENODEV;
1198
1199         if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
1200                 return -EOPNOTSUPP;
1201
1202         rndis_dev = ndev->extension;
1203         if (indir) {
1204                 for (i = 0; i < ITAB_NUM; i++)
1205                         if (indir[i] >= VRSS_CHANNEL_MAX)
1206                                 return -EINVAL;
1207
1208                 for (i = 0; i < ITAB_NUM; i++)
1209                         rndis_dev->ind_table[i] = indir[i];
1210         }
1211
1212         if (!key) {
1213                 if (!indir)
1214                         return 0;
1215
1216                 key = rndis_dev->rss_key;
1217         }
1218
1219         return rndis_filter_set_rss_param(rndis_dev, key, ndev->num_chn);
1220 }
1221
1222 static const struct ethtool_ops ethtool_ops = {
1223         .get_drvinfo    = netvsc_get_drvinfo,
1224         .get_link       = ethtool_op_get_link,
1225         .get_ethtool_stats = netvsc_get_ethtool_stats,
1226         .get_sset_count = netvsc_get_sset_count,
1227         .get_strings    = netvsc_get_strings,
1228         .get_channels   = netvsc_get_channels,
1229         .set_channels   = netvsc_set_channels,
1230         .get_ts_info    = ethtool_op_get_ts_info,
1231         .get_rxnfc      = netvsc_get_rxnfc,
1232         .get_rxfh_key_size = netvsc_get_rxfh_key_size,
1233         .get_rxfh_indir_size = netvsc_rss_indir_size,
1234         .get_rxfh       = netvsc_get_rxfh,
1235         .set_rxfh       = netvsc_set_rxfh,
1236         .get_link_ksettings = netvsc_get_link_ksettings,
1237         .set_link_ksettings = netvsc_set_link_ksettings,
1238 };
1239
1240 static const struct net_device_ops device_ops = {
1241         .ndo_open =                     netvsc_open,
1242         .ndo_stop =                     netvsc_close,
1243         .ndo_start_xmit =               netvsc_start_xmit,
1244         .ndo_set_rx_mode =              netvsc_set_multicast_list,
1245         .ndo_change_mtu =               netvsc_change_mtu,
1246         .ndo_validate_addr =            eth_validate_addr,
1247         .ndo_set_mac_address =          netvsc_set_mac_addr,
1248         .ndo_select_queue =             netvsc_select_queue,
1249         .ndo_get_stats64 =              netvsc_get_stats64,
1250 #ifdef CONFIG_NET_POLL_CONTROLLER
1251         .ndo_poll_controller =          netvsc_poll_controller,
1252 #endif
1253 };
1254
1255 /*
1256  * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
1257  * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
1258  * present send GARP packet to network peers with netif_notify_peers().
1259  */
1260 static void netvsc_link_change(struct work_struct *w)
1261 {
1262         struct net_device_context *ndev_ctx =
1263                 container_of(w, struct net_device_context, dwork.work);
1264         struct hv_device *device_obj = ndev_ctx->device_ctx;
1265         struct net_device *net = hv_get_drvdata(device_obj);
1266         struct netvsc_device *net_device;
1267         struct rndis_device *rdev;
1268         struct netvsc_reconfig *event = NULL;
1269         bool notify = false, reschedule = false;
1270         unsigned long flags, next_reconfig, delay;
1271
1272         rtnl_lock();
1273         net_device = rtnl_dereference(ndev_ctx->nvdev);
1274         if (!net_device)
1275                 goto out_unlock;
1276
1277         rdev = net_device->extension;
1278
1279         next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
1280         if (time_is_after_jiffies(next_reconfig)) {
1281                 /* link_watch only sends one notification with current state
1282                  * per second, avoid doing reconfig more frequently. Handle
1283                  * wrap around.
1284                  */
1285                 delay = next_reconfig - jiffies;
1286                 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
1287                 schedule_delayed_work(&ndev_ctx->dwork, delay);
1288                 goto out_unlock;
1289         }
1290         ndev_ctx->last_reconfig = jiffies;
1291
1292         spin_lock_irqsave(&ndev_ctx->lock, flags);
1293         if (!list_empty(&ndev_ctx->reconfig_events)) {
1294                 event = list_first_entry(&ndev_ctx->reconfig_events,
1295                                          struct netvsc_reconfig, list);
1296                 list_del(&event->list);
1297                 reschedule = !list_empty(&ndev_ctx->reconfig_events);
1298         }
1299         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1300
1301         if (!event)
1302                 goto out_unlock;
1303
1304         switch (event->event) {
1305                 /* Only the following events are possible due to the check in
1306                  * netvsc_linkstatus_callback()
1307                  */
1308         case RNDIS_STATUS_MEDIA_CONNECT:
1309                 if (rdev->link_state) {
1310                         rdev->link_state = false;
1311                         if (!ndev_ctx->datapath)
1312                                 netif_carrier_on(net);
1313                         netif_tx_wake_all_queues(net);
1314                 } else {
1315                         notify = true;
1316                 }
1317                 kfree(event);
1318                 break;
1319         case RNDIS_STATUS_MEDIA_DISCONNECT:
1320                 if (!rdev->link_state) {
1321                         rdev->link_state = true;
1322                         netif_carrier_off(net);
1323                         netif_tx_stop_all_queues(net);
1324                 }
1325                 kfree(event);
1326                 break;
1327         case RNDIS_STATUS_NETWORK_CHANGE:
1328                 /* Only makes sense if carrier is present */
1329                 if (!rdev->link_state) {
1330                         rdev->link_state = true;
1331                         netif_carrier_off(net);
1332                         netif_tx_stop_all_queues(net);
1333                         event->event = RNDIS_STATUS_MEDIA_CONNECT;
1334                         spin_lock_irqsave(&ndev_ctx->lock, flags);
1335                         list_add(&event->list, &ndev_ctx->reconfig_events);
1336                         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1337                         reschedule = true;
1338                 }
1339                 break;
1340         }
1341
1342         rtnl_unlock();
1343
1344         if (notify)
1345                 netdev_notify_peers(net);
1346
1347         /* link_watch only sends one notification with current state per
1348          * second, handle next reconfig event in 2 seconds.
1349          */
1350         if (reschedule)
1351                 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1352
1353         return;
1354
1355 out_unlock:
1356         rtnl_unlock();
1357 }
1358
1359 static struct net_device *get_netvsc_bymac(const u8 *mac)
1360 {
1361         struct net_device *dev;
1362
1363         ASSERT_RTNL();
1364
1365         for_each_netdev(&init_net, dev) {
1366                 if (dev->netdev_ops != &device_ops)
1367                         continue;       /* not a netvsc device */
1368
1369                 if (ether_addr_equal(mac, dev->perm_addr))
1370                         return dev;
1371         }
1372
1373         return NULL;
1374 }
1375
1376 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
1377 {
1378         struct net_device *dev;
1379
1380         ASSERT_RTNL();
1381
1382         for_each_netdev(&init_net, dev) {
1383                 struct net_device_context *net_device_ctx;
1384
1385                 if (dev->netdev_ops != &device_ops)
1386                         continue;       /* not a netvsc device */
1387
1388                 net_device_ctx = netdev_priv(dev);
1389                 if (net_device_ctx->nvdev == NULL)
1390                         continue;       /* device is removed */
1391
1392                 if (rtnl_dereference(net_device_ctx->vf_netdev) == vf_netdev)
1393                         return dev;     /* a match */
1394         }
1395
1396         return NULL;
1397 }
1398
1399 static int netvsc_register_vf(struct net_device *vf_netdev)
1400 {
1401         struct net_device *ndev;
1402         struct net_device_context *net_device_ctx;
1403         struct netvsc_device *netvsc_dev;
1404
1405         if (vf_netdev->addr_len != ETH_ALEN)
1406                 return NOTIFY_DONE;
1407
1408         /*
1409          * We will use the MAC address to locate the synthetic interface to
1410          * associate with the VF interface. If we don't find a matching
1411          * synthetic interface, move on.
1412          */
1413         ndev = get_netvsc_bymac(vf_netdev->perm_addr);
1414         if (!ndev)
1415                 return NOTIFY_DONE;
1416
1417         net_device_ctx = netdev_priv(ndev);
1418         netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
1419         if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
1420                 return NOTIFY_DONE;
1421
1422         netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
1423         /*
1424          * Take a reference on the module.
1425          */
1426         try_module_get(THIS_MODULE);
1427
1428         dev_hold(vf_netdev);
1429         rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
1430         return NOTIFY_OK;
1431 }
1432
1433 static int netvsc_vf_up(struct net_device *vf_netdev)
1434 {
1435         struct net_device *ndev;
1436         struct netvsc_device *netvsc_dev;
1437         struct net_device_context *net_device_ctx;
1438
1439         ndev = get_netvsc_byref(vf_netdev);
1440         if (!ndev)
1441                 return NOTIFY_DONE;
1442
1443         net_device_ctx = netdev_priv(ndev);
1444         netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
1445
1446         netdev_info(ndev, "VF up: %s\n", vf_netdev->name);
1447
1448         /*
1449          * Open the device before switching data path.
1450          */
1451         rndis_filter_open(netvsc_dev);
1452
1453         /*
1454          * notify the host to switch the data path.
1455          */
1456         netvsc_switch_datapath(ndev, true);
1457         netdev_info(ndev, "Data path switched to VF: %s\n", vf_netdev->name);
1458
1459         netif_carrier_off(ndev);
1460
1461         /* Now notify peers through VF device. */
1462         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, vf_netdev);
1463
1464         return NOTIFY_OK;
1465 }
1466
1467 static int netvsc_vf_down(struct net_device *vf_netdev)
1468 {
1469         struct net_device *ndev;
1470         struct netvsc_device *netvsc_dev;
1471         struct net_device_context *net_device_ctx;
1472
1473         ndev = get_netvsc_byref(vf_netdev);
1474         if (!ndev)
1475                 return NOTIFY_DONE;
1476
1477         net_device_ctx = netdev_priv(ndev);
1478         netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
1479
1480         netdev_info(ndev, "VF down: %s\n", vf_netdev->name);
1481         netvsc_switch_datapath(ndev, false);
1482         netdev_info(ndev, "Data path switched from VF: %s\n", vf_netdev->name);
1483         rndis_filter_close(netvsc_dev);
1484         netif_carrier_on(ndev);
1485
1486         /* Now notify peers through netvsc device. */
1487         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, ndev);
1488
1489         return NOTIFY_OK;
1490 }
1491
1492 static int netvsc_unregister_vf(struct net_device *vf_netdev)
1493 {
1494         struct net_device *ndev;
1495         struct net_device_context *net_device_ctx;
1496
1497         ndev = get_netvsc_byref(vf_netdev);
1498         if (!ndev)
1499                 return NOTIFY_DONE;
1500
1501         net_device_ctx = netdev_priv(ndev);
1502
1503         netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
1504
1505         RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
1506         dev_put(vf_netdev);
1507         module_put(THIS_MODULE);
1508         return NOTIFY_OK;
1509 }
1510
1511 static int netvsc_probe(struct hv_device *dev,
1512                         const struct hv_vmbus_device_id *dev_id)
1513 {
1514         struct net_device *net = NULL;
1515         struct net_device_context *net_device_ctx;
1516         struct netvsc_device_info device_info;
1517         struct netvsc_device *nvdev;
1518         int ret;
1519
1520         net = alloc_etherdev_mq(sizeof(struct net_device_context),
1521                                 VRSS_CHANNEL_MAX);
1522         if (!net)
1523                 return -ENOMEM;
1524
1525         netif_carrier_off(net);
1526
1527         netvsc_init_settings(net);
1528
1529         net_device_ctx = netdev_priv(net);
1530         net_device_ctx->device_ctx = dev;
1531         net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
1532         if (netif_msg_probe(net_device_ctx))
1533                 netdev_dbg(net, "netvsc msg_enable: %d\n",
1534                            net_device_ctx->msg_enable);
1535
1536         hv_set_drvdata(dev, net);
1537
1538         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
1539
1540         spin_lock_init(&net_device_ctx->lock);
1541         INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
1542
1543         net->netdev_ops = &device_ops;
1544         net->ethtool_ops = &ethtool_ops;
1545         SET_NETDEV_DEV(net, &dev->device);
1546
1547         /* We always need headroom for rndis header */
1548         net->needed_headroom = RNDIS_AND_PPI_SIZE;
1549
1550         /* Notify the netvsc driver of the new device */
1551         memset(&device_info, 0, sizeof(device_info));
1552         device_info.ring_size = ring_size;
1553         device_info.num_chn = VRSS_CHANNEL_DEFAULT;
1554         ret = rndis_filter_device_add(dev, &device_info);
1555         if (ret != 0) {
1556                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
1557                 free_netdev(net);
1558                 hv_set_drvdata(dev, NULL);
1559                 return ret;
1560         }
1561         memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
1562
1563         /* hw_features computed in rndis_filter_device_add */
1564         net->features = net->hw_features |
1565                 NETIF_F_HIGHDMA | NETIF_F_SG |
1566                 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
1567         net->vlan_features = net->features;
1568
1569         /* RCU not necessary here, device not registered */
1570         nvdev = net_device_ctx->nvdev;
1571         netif_set_real_num_tx_queues(net, nvdev->num_chn);
1572         netif_set_real_num_rx_queues(net, nvdev->num_chn);
1573
1574         /* MTU range: 68 - 1500 or 65521 */
1575         net->min_mtu = NETVSC_MTU_MIN;
1576         if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
1577                 net->max_mtu = NETVSC_MTU - ETH_HLEN;
1578         else
1579                 net->max_mtu = ETH_DATA_LEN;
1580
1581         ret = register_netdev(net);
1582         if (ret != 0) {
1583                 pr_err("Unable to register netdev.\n");
1584                 rndis_filter_device_remove(dev, nvdev);
1585                 free_netdev(net);
1586         }
1587
1588         return ret;
1589 }
1590
1591 static int netvsc_remove(struct hv_device *dev)
1592 {
1593         struct net_device *net;
1594         struct net_device_context *ndev_ctx;
1595
1596         net = hv_get_drvdata(dev);
1597
1598         if (net == NULL) {
1599                 dev_err(&dev->device, "No net device to remove\n");
1600                 return 0;
1601         }
1602
1603         ndev_ctx = netdev_priv(net);
1604
1605         netif_device_detach(net);
1606
1607         cancel_delayed_work_sync(&ndev_ctx->dwork);
1608
1609         /*
1610          * Call to the vsc driver to let it know that the device is being
1611          * removed. Also blocks mtu and channel changes.
1612          */
1613         rtnl_lock();
1614         rndis_filter_device_remove(dev, ndev_ctx->nvdev);
1615         rtnl_unlock();
1616
1617         unregister_netdev(net);
1618
1619         hv_set_drvdata(dev, NULL);
1620
1621         free_netdev(net);
1622         return 0;
1623 }
1624
1625 static const struct hv_vmbus_device_id id_table[] = {
1626         /* Network guid */
1627         { HV_NIC_GUID, },
1628         { },
1629 };
1630
1631 MODULE_DEVICE_TABLE(vmbus, id_table);
1632
1633 /* The one and only one */
1634 static struct  hv_driver netvsc_drv = {
1635         .name = KBUILD_MODNAME,
1636         .id_table = id_table,
1637         .probe = netvsc_probe,
1638         .remove = netvsc_remove,
1639 };
1640
1641 /*
1642  * On Hyper-V, every VF interface is matched with a corresponding
1643  * synthetic interface. The synthetic interface is presented first
1644  * to the guest. When the corresponding VF instance is registered,
1645  * we will take care of switching the data path.
1646  */
1647 static int netvsc_netdev_event(struct notifier_block *this,
1648                                unsigned long event, void *ptr)
1649 {
1650         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
1651
1652         /* Skip our own events */
1653         if (event_dev->netdev_ops == &device_ops)
1654                 return NOTIFY_DONE;
1655
1656         /* Avoid non-Ethernet type devices */
1657         if (event_dev->type != ARPHRD_ETHER)
1658                 return NOTIFY_DONE;
1659
1660         /* Avoid Vlan dev with same MAC registering as VF */
1661         if (is_vlan_dev(event_dev))
1662                 return NOTIFY_DONE;
1663
1664         /* Avoid Bonding master dev with same MAC registering as VF */
1665         if ((event_dev->priv_flags & IFF_BONDING) &&
1666             (event_dev->flags & IFF_MASTER))
1667                 return NOTIFY_DONE;
1668
1669         switch (event) {
1670         case NETDEV_REGISTER:
1671                 return netvsc_register_vf(event_dev);
1672         case NETDEV_UNREGISTER:
1673                 return netvsc_unregister_vf(event_dev);
1674         case NETDEV_UP:
1675                 return netvsc_vf_up(event_dev);
1676         case NETDEV_DOWN:
1677                 return netvsc_vf_down(event_dev);
1678         default:
1679                 return NOTIFY_DONE;
1680         }
1681 }
1682
1683 static struct notifier_block netvsc_netdev_notifier = {
1684         .notifier_call = netvsc_netdev_event,
1685 };
1686
1687 static void __exit netvsc_drv_exit(void)
1688 {
1689         unregister_netdevice_notifier(&netvsc_netdev_notifier);
1690         vmbus_driver_unregister(&netvsc_drv);
1691 }
1692
1693 static int __init netvsc_drv_init(void)
1694 {
1695         int ret;
1696
1697         if (ring_size < RING_SIZE_MIN) {
1698                 ring_size = RING_SIZE_MIN;
1699                 pr_info("Increased ring_size to %d (min allowed)\n",
1700                         ring_size);
1701         }
1702         ret = vmbus_driver_register(&netvsc_drv);
1703
1704         if (ret)
1705                 return ret;
1706
1707         register_netdevice_notifier(&netvsc_netdev_notifier);
1708         return 0;
1709 }
1710
1711 MODULE_LICENSE("GPL");
1712 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
1713
1714 module_init(netvsc_drv_init);
1715 module_exit(netvsc_drv_exit);