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
  41 #include "hyperv_net.h"
  42
  43 struct net_device_context {
  44         /* point back to our device context */
  45         struct hv_device *device_ctx;
  46         struct delayed_work dwork;
  47         struct work_struct work;
  48 };
  49
  50 #define RING_SIZE_MIN 64
  51 static int ring_size = 128;
  52 module_param(ring_size, int, S_IRUGO);
  53 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
  54
  55 static void do_set_multicast(struct work_struct *w)
  56 {
  57         struct net_device_context *ndevctx =
  58                 container_of(w, struct net_device_context, work);
  59         struct netvsc_device *nvdev;
  60         struct rndis_device *rdev;
  61
  62         nvdev = hv_get_drvdata(ndevctx->device_ctx);
  63         if (nvdev == NULL || nvdev->ndev == NULL)
  64                 return;
  65
  66         rdev = nvdev->extension;
  67         if (rdev == NULL)
  68                 return;
  69
  70         if (nvdev->ndev->flags & IFF_PROMISC)
  71                 rndis_filter_set_packet_filter(rdev,
  72                         NDIS_PACKET_TYPE_PROMISCUOUS);
  73         else
  74                 rndis_filter_set_packet_filter(rdev,
  75                         NDIS_PACKET_TYPE_BROADCAST |
  76                         NDIS_PACKET_TYPE_ALL_MULTICAST |
  77                         NDIS_PACKET_TYPE_DIRECTED);
  78 }
  79
  80 static void netvsc_set_multicast_list(struct net_device *net)
  81 {
  82         struct net_device_context *net_device_ctx = netdev_priv(net);
  83
  84         schedule_work(&net_device_ctx->work);
  85 }
  86
  87 static int netvsc_open(struct net_device *net)
  88 {
  89         struct net_device_context *net_device_ctx = netdev_priv(net);
  90         struct hv_device *device_obj = net_device_ctx->device_ctx;
  91         struct netvsc_device *nvdev;
  92         struct rndis_device *rdev;
  93         int ret = 0;
  94
  95         netif_carrier_off(net);
  96
  97         /* Open up the device */
  98         ret = rndis_filter_open(device_obj);
  99         if (ret != 0) {
 100                 netdev_err(net, "unable to open device (ret %d).\n", ret);
 101                 return ret;
 102         }
 103
 104         netif_tx_start_all_queues(net);
 105
 106         nvdev = hv_get_drvdata(device_obj);
 107         rdev = nvdev->extension;
 108         if (!rdev->link_state)
 109                 netif_carrier_on(net);
 110
 111         return ret;
 112 }
 113
 114 static int netvsc_close(struct net_device *net)
 115 {
 116         struct net_device_context *net_device_ctx = netdev_priv(net);
 117         struct hv_device *device_obj = net_device_ctx->device_ctx;
 118         int ret;
 119
 120         netif_tx_disable(net);
 121
 122         /* Make sure netvsc_set_multicast_list doesn't re-enable filter! */
 123         cancel_work_sync(&net_device_ctx->work);
 124         ret = rndis_filter_close(device_obj);
 125         if (ret != 0)
 126                 netdev_err(net, "unable to close device (ret %d).\n", ret);
 127
 128         return ret;
 129 }
 130
 131 static void *init_ppi_data(struct rndis_message *msg, u32 ppi_size,
 132                                 int pkt_type)
 133 {
 134         struct rndis_packet *rndis_pkt;
 135         struct rndis_per_packet_info *ppi;
 136
 137         rndis_pkt = &msg->msg.pkt;
 138         rndis_pkt->data_offset += ppi_size;
 139
 140         ppi = (struct rndis_per_packet_info *)((void *)rndis_pkt +
 141                 rndis_pkt->per_pkt_info_offset + rndis_pkt->per_pkt_info_len);
 142
 143         ppi->size = ppi_size;
 144         ppi->type = pkt_type;
 145         ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
 146
 147         rndis_pkt->per_pkt_info_len += ppi_size;
 148
 149         return ppi;
 150 }
 151
 152 union sub_key {
 153         u64 k;
 154         struct {
 155                 u8 pad[3];
 156                 u8 kb;
 157                 u32 ka;
 158         };
 159 };
 160
 161 /* Toeplitz hash function
 162  * data: network byte order
 163  * return: host byte order
 164  */
 165 static u32 comp_hash(u8 *key, int klen, u8 *data, int dlen)
 166 {
 167         union sub_key subk;
 168         int k_next = 4;
 169         u8 dt;
 170         int i, j;
 171         u32 ret = 0;
 172
 173         subk.k = 0;
 174         subk.ka = ntohl(*(u32 *)key);
 175
 176         for (i = 0; i < dlen; i++) {
 177                 subk.kb = key[k_next];
 178                 k_next = (k_next + 1) % klen;
 179                 dt = data[i];
 180                 for (j = 0; j < 8; j++) {
 181                         if (dt & 0x80)
 182                                 ret ^= subk.ka;
 183                         dt <<= 1;
 184                         subk.k <<= 1;
 185                 }
 186         }
 187
 188         return ret;
 189 }
 190
 191 static bool netvsc_set_hash(u32 *hash, struct sk_buff *skb)
 192 {
 193         struct iphdr *iphdr;
 194         int data_len;
 195         bool ret = false;
 196
 197         if (eth_hdr(skb)->h_proto != htons(ETH_P_IP))
 198                 return false;
 199
 200         iphdr = ip_hdr(skb);
 201
 202         if (iphdr->version == 4) {
 203                 if (iphdr->protocol == IPPROTO_TCP)
 204                         data_len = 12;
 205                 else
 206                         data_len = 8;
 207                 *hash = comp_hash(netvsc_hash_key, HASH_KEYLEN,
 208                                   (u8 *)&iphdr->saddr, data_len);
 209                 ret = true;
 210         }
 211
 212         return ret;
 213 }
 214
 215 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
 216                         void *accel_priv, select_queue_fallback_t fallback)
 217 {
 218         struct net_device_context *net_device_ctx = netdev_priv(ndev);
 219         struct hv_device *hdev =  net_device_ctx->device_ctx;
 220         struct netvsc_device *nvsc_dev = hv_get_drvdata(hdev);
 221         u32 hash;
 222         u16 q_idx = 0;
 223
 224         if (nvsc_dev == NULL || ndev->real_num_tx_queues <= 1)
 225                 return 0;
 226
 227         if (netvsc_set_hash(&hash, skb))
 228                 q_idx = nvsc_dev->send_table[hash % VRSS_SEND_TAB_SIZE] %
 229                         ndev->real_num_tx_queues;
 230
 231         return q_idx;
 232 }
 233
 234 static void netvsc_xmit_completion(void *context)
 235 {
 236         struct hv_netvsc_packet *packet = (struct hv_netvsc_packet *)context;
 237         struct sk_buff *skb = (struct sk_buff *)
 238                 (unsigned long)packet->send_completion_tid;
 239         u32 index = packet->send_buf_index;
 240
 241         kfree(packet);
 242
 243         if (skb && (index == NETVSC_INVALID_INDEX))
 244                 dev_kfree_skb_any(skb);
 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_page_buffer *pb)
 283 {
 284         u32 slots_used = 0;
 285         char *data = skb->data;
 286         int frags = skb_shinfo(skb)->nr_frags;
 287         int i;
 288
 289         /* The packet is laid out thus:
 290          * 1. hdr
 291          * 2. skb linear data
 292          * 3. skb fragment data
 293          */
 294         if (hdr != NULL)
 295                 slots_used += fill_pg_buf(virt_to_page(hdr),
 296                                         offset_in_page(hdr),
 297                                         len, &pb[slots_used]);
 298
 299         slots_used += fill_pg_buf(virt_to_page(data),
 300                                 offset_in_page(data),
 301                                 skb_headlen(skb), &pb[slots_used]);
 302
 303         for (i = 0; i < frags; i++) {
 304                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 305
 306                 slots_used += fill_pg_buf(skb_frag_page(frag),
 307                                         frag->page_offset,
 308                                         skb_frag_size(frag), &pb[slots_used]);
 309         }
 310         return slots_used;
 311 }
 312
 313 static int count_skb_frag_slots(struct sk_buff *skb)
 314 {
 315         int i, frags = skb_shinfo(skb)->nr_frags;
 316         int pages = 0;
 317
 318         for (i = 0; i < frags; i++) {
 319                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 320                 unsigned long size = skb_frag_size(frag);
 321                 unsigned long offset = frag->page_offset;
 322
 323                 /* Skip unused frames from start of page */
 324                 offset &= ~PAGE_MASK;
 325                 pages += PFN_UP(offset + size);
 326         }
 327         return pages;
 328 }
 329
 330 static int netvsc_get_slots(struct sk_buff *skb)
 331 {
 332         char *data = skb->data;
 333         unsigned int offset = offset_in_page(data);
 334         unsigned int len = skb_headlen(skb);
 335         int slots;
 336         int frag_slots;
 337
 338         slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
 339         frag_slots = count_skb_frag_slots(skb);
 340         return slots + frag_slots;
 341 }
 342
 343 static u32 get_net_transport_info(struct sk_buff *skb, u32 *trans_off)
 344 {
 345         u32 ret_val = TRANSPORT_INFO_NOT_IP;
 346
 347         if ((eth_hdr(skb)->h_proto != htons(ETH_P_IP)) &&
 348                 (eth_hdr(skb)->h_proto != htons(ETH_P_IPV6))) {
 349                 goto not_ip;
 350         }
 351
 352         *trans_off = skb_transport_offset(skb);
 353
 354         if ((eth_hdr(skb)->h_proto == htons(ETH_P_IP))) {
 355                 struct iphdr *iphdr = ip_hdr(skb);
 356
 357                 if (iphdr->protocol == IPPROTO_TCP)
 358                         ret_val = TRANSPORT_INFO_IPV4_TCP;
 359                 else if (iphdr->protocol == IPPROTO_UDP)
 360                         ret_val = TRANSPORT_INFO_IPV4_UDP;
 361         } else {
 362                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
 363                         ret_val = TRANSPORT_INFO_IPV6_TCP;
 364                 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
 365                         ret_val = TRANSPORT_INFO_IPV6_UDP;
 366         }
 367
 368 not_ip:
 369         return ret_val;
 370 }
 371
 372 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
 373 {
 374         struct net_device_context *net_device_ctx = netdev_priv(net);
 375         struct hv_netvsc_packet *packet;
 376         int ret;
 377         unsigned int num_data_pgs;
 378         struct rndis_message *rndis_msg;
 379         struct rndis_packet *rndis_pkt;
 380         u32 rndis_msg_size;
 381         bool isvlan;
 382         struct rndis_per_packet_info *ppi;
 383         struct ndis_tcp_ip_checksum_info *csum_info;
 384         struct ndis_tcp_lso_info *lso_info;
 385         int  hdr_offset;
 386         u32 net_trans_info;
 387
 388
 389         /* We will atmost need two pages to describe the rndis
 390          * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
 391          * of pages in a single packet.
 392          */
 393         num_data_pgs = netvsc_get_slots(skb) + 2;
 394         if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
 395                 netdev_err(net, "Packet too big: %u\n", skb->len);
 396                 dev_kfree_skb(skb);
 397                 net->stats.tx_dropped++;
 398                 return NETDEV_TX_OK;
 399         }
 400
 401         /* Allocate a netvsc packet based on # of frags. */
 402         packet = kzalloc(sizeof(struct hv_netvsc_packet) +
 403                          (num_data_pgs * sizeof(struct hv_page_buffer)) +
 404                          sizeof(struct rndis_message) +
 405                          NDIS_VLAN_PPI_SIZE +
 406                          NDIS_CSUM_PPI_SIZE +
 407                          NDIS_LSO_PPI_SIZE, GFP_ATOMIC);
 408         if (!packet) {
 409                 /* out of memory, drop packet */
 410                 netdev_err(net, "unable to allocate hv_netvsc_packet\n");
 411
 412                 dev_kfree_skb(skb);
 413                 net->stats.tx_dropped++;
 414                 return NETDEV_TX_OK;
 415         }
 416
 417         packet->vlan_tci = skb->vlan_tci;
 418
 419         packet->q_idx = skb_get_queue_mapping(skb);
 420
 421         packet->is_data_pkt = true;
 422         packet->total_data_buflen = skb->len;
 423
 424         packet->rndis_msg = (struct rndis_message *)((unsigned long)packet +
 425                                 sizeof(struct hv_netvsc_packet) +
 426                                 (num_data_pgs * sizeof(struct hv_page_buffer)));
 427
 428         /* Set the completion routine */
 429         packet->send_completion = netvsc_xmit_completion;
 430         packet->send_completion_ctx = packet;
 431         packet->send_completion_tid = (unsigned long)skb;
 432
 433         isvlan = packet->vlan_tci & VLAN_TAG_PRESENT;
 434
 435         /* Add the rndis header */
 436         rndis_msg = packet->rndis_msg;
 437         rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
 438         rndis_msg->msg_len = packet->total_data_buflen;
 439         rndis_pkt = &rndis_msg->msg.pkt;
 440         rndis_pkt->data_offset = sizeof(struct rndis_packet);
 441         rndis_pkt->data_len = packet->total_data_buflen;
 442         rndis_pkt->per_pkt_info_offset = sizeof(struct rndis_packet);
 443
 444         rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
 445
 446         if (isvlan) {
 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 = packet->vlan_tci & VLAN_VID_MASK;
 455                 vlan->pri = (packet->vlan_tci & VLAN_PRIO_MASK) >>
 456                                 VLAN_PRIO_SHIFT;
 457         }
 458
 459         net_trans_info = get_net_transport_info(skb, &hdr_offset);
 460         if (net_trans_info == TRANSPORT_INFO_NOT_IP)
 461                 goto do_send;
 462
 463         /*
 464          * Setup the sendside checksum offload only if this is not a
 465          * GSO packet.
 466          */
 467         if (skb_is_gso(skb))
 468                 goto do_lso;
 469
 470         rndis_msg_size += NDIS_CSUM_PPI_SIZE;
 471         ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
 472                             TCPIP_CHKSUM_PKTINFO);
 473
 474         csum_info = (struct ndis_tcp_ip_checksum_info *)((void *)ppi +
 475                         ppi->ppi_offset);
 476
 477         if (net_trans_info & (INFO_IPV4 << 16))
 478                 csum_info->transmit.is_ipv4 = 1;
 479         else
 480                 csum_info->transmit.is_ipv6 = 1;
 481
 482         if (net_trans_info & INFO_TCP) {
 483                 csum_info->transmit.tcp_checksum = 1;
 484                 csum_info->transmit.tcp_header_offset = hdr_offset;
 485         } else if (net_trans_info & INFO_UDP) {
 486                 /* UDP checksum offload is not supported on ws2008r2.
 487                  * Furthermore, on ws2012 and ws2012r2, there are some
 488                  * issues with udp checksum offload from Linux guests.
 489                  * (these are host issues).
 490                  * For now compute the checksum here.
 491                  */
 492                 struct udphdr *uh;
 493                 u16 udp_len;
 494
 495                 ret = skb_cow_head(skb, 0);
 496                 if (ret)
 497                         goto drop;
 498
 499                 uh = udp_hdr(skb);
 500                 udp_len = ntohs(uh->len);
 501                 uh->check = 0;
 502                 uh->check = csum_tcpudp_magic(ip_hdr(skb)->saddr,
 503                                               ip_hdr(skb)->daddr,
 504                                               udp_len, IPPROTO_UDP,
 505                                               csum_partial(uh, udp_len, 0));
 506                 if (uh->check == 0)
 507                         uh->check = CSUM_MANGLED_0;
 508
 509                 csum_info->transmit.udp_checksum = 0;
 510         }
 511         goto do_send;
 512
 513 do_lso:
 514         rndis_msg_size += NDIS_LSO_PPI_SIZE;
 515         ppi = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
 516                             TCP_LARGESEND_PKTINFO);
 517
 518         lso_info = (struct ndis_tcp_lso_info *)((void *)ppi +
 519                         ppi->ppi_offset);
 520
 521         lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
 522         if (net_trans_info & (INFO_IPV4 << 16)) {
 523                 lso_info->lso_v2_transmit.ip_version =
 524                         NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
 525                 ip_hdr(skb)->tot_len = 0;
 526                 ip_hdr(skb)->check = 0;
 527                 tcp_hdr(skb)->check =
 528                 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
 529                                    ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 530         } else {
 531                 lso_info->lso_v2_transmit.ip_version =
 532                         NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
 533                 ipv6_hdr(skb)->payload_len = 0;
 534                 tcp_hdr(skb)->check =
 535                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 536                                 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 537         }
 538         lso_info->lso_v2_transmit.tcp_header_offset = hdr_offset;
 539         lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
 540
 541 do_send:
 542         /* Start filling in the page buffers with the rndis hdr */
 543         rndis_msg->msg_len += rndis_msg_size;
 544         packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
 545                                         skb, &packet->page_buf[0]);
 546
 547         ret = netvsc_send(net_device_ctx->device_ctx, packet);
 548
 549 drop:
 550         if (ret == 0) {
 551                 net->stats.tx_bytes += skb->len;
 552                 net->stats.tx_packets++;
 553         } else {
 554                 kfree(packet);
 555                 if (ret != -EAGAIN) {
 556                         dev_kfree_skb_any(skb);
 557                         net->stats.tx_dropped++;
 558                 }
 559         }
 560
 561         return (ret == -EAGAIN) ? NETDEV_TX_BUSY : NETDEV_TX_OK;
 562 }
 563
 564 /*
 565  * netvsc_linkstatus_callback - Link up/down notification
 566  */
 567 void netvsc_linkstatus_callback(struct hv_device *device_obj,
 568                                        unsigned int status)
 569 {
 570         struct net_device *net;
 571         struct net_device_context *ndev_ctx;
 572         struct netvsc_device *net_device;
 573         struct rndis_device *rdev;
 574
 575         net_device = hv_get_drvdata(device_obj);
 576         rdev = net_device->extension;
 577
 578         rdev->link_state = status != 1;
 579
 580         net = net_device->ndev;
 581
 582         if (!net || net->reg_state != NETREG_REGISTERED)
 583                 return;
 584
 585         ndev_ctx = netdev_priv(net);
 586         if (status == 1) {
 587                 schedule_delayed_work(&ndev_ctx->dwork, 0);
 588                 schedule_delayed_work(&ndev_ctx->dwork, msecs_to_jiffies(20));
 589         } else {
 590                 schedule_delayed_work(&ndev_ctx->dwork, 0);
 591         }
 592 }
 593
 594 /*
 595  * netvsc_recv_callback -  Callback when we receive a packet from the
 596  * "wire" on the specified device.
 597  */
 598 int netvsc_recv_callback(struct hv_device *device_obj,
 599                                 struct hv_netvsc_packet *packet,
 600                                 struct ndis_tcp_ip_checksum_info *csum_info)
 601 {
 602         struct net_device *net;
 603         struct sk_buff *skb;
 604
 605         net = ((struct netvsc_device *)hv_get_drvdata(device_obj))->ndev;
 606         if (!net || net->reg_state != NETREG_REGISTERED) {
 607                 packet->status = NVSP_STAT_FAIL;
 608                 return 0;
 609         }
 610
 611         /* Allocate a skb - TODO direct I/O to pages? */
 612         skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
 613         if (unlikely(!skb)) {
 614                 ++net->stats.rx_dropped;
 615                 packet->status = NVSP_STAT_FAIL;
 616                 return 0;
 617         }
 618
 619         /*
 620          * Copy to skb. This copy is needed here since the memory pointed by
 621          * hv_netvsc_packet cannot be deallocated
 622          */
 623         memcpy(skb_put(skb, packet->total_data_buflen), packet->data,
 624                 packet->total_data_buflen);
 625
 626         skb->protocol = eth_type_trans(skb, net);
 627         if (csum_info) {
 628                 /* We only look at the IP checksum here.
 629                  * Should we be dropping the packet if checksum
 630                  * failed? How do we deal with other checksums - TCP/UDP?
 631                  */
 632                 if (csum_info->receive.ip_checksum_succeeded)
 633                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 634                 else
 635                         skb->ip_summed = CHECKSUM_NONE;
 636         }
 637
 638         if (packet->vlan_tci & VLAN_TAG_PRESENT)
 639                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 640                                        packet->vlan_tci);
 641
 642         skb_record_rx_queue(skb, packet->channel->
 643                             offermsg.offer.sub_channel_index %
 644                             net->real_num_rx_queues);
 645
 646         net->stats.rx_packets++;
 647         net->stats.rx_bytes += packet->total_data_buflen;
 648
 649         /*
 650          * Pass the skb back up. Network stack will deallocate the skb when it
 651          * is done.
 652          * TODO - use NAPI?
 653          */
 654         netif_rx(skb);
 655
 656         return 0;
 657 }
 658
 659 static void netvsc_get_drvinfo(struct net_device *net,
 660                                struct ethtool_drvinfo *info)
 661 {
 662         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
 663         strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
 664 }
 665
 666 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
 667 {
 668         struct net_device_context *ndevctx = netdev_priv(ndev);
 669         struct hv_device *hdev =  ndevctx->device_ctx;
 670         struct netvsc_device *nvdev = hv_get_drvdata(hdev);
 671         struct netvsc_device_info device_info;
 672         int limit = ETH_DATA_LEN;
 673
 674         if (nvdev == NULL || nvdev->destroy)
 675                 return -ENODEV;
 676
 677         if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
 678                 limit = NETVSC_MTU;
 679
 680         if (mtu < 68 || mtu > limit)
 681                 return -EINVAL;
 682
 683         nvdev->start_remove = true;
 684         cancel_work_sync(&ndevctx->work);
 685         netif_tx_disable(ndev);
 686         rndis_filter_device_remove(hdev);
 687
 688         ndev->mtu = mtu;
 689
 690         ndevctx->device_ctx = hdev;
 691         hv_set_drvdata(hdev, ndev);
 692         device_info.ring_size = ring_size;
 693         rndis_filter_device_add(hdev, &device_info);
 694         netif_tx_wake_all_queues(ndev);
 695
 696         return 0;
 697 }
 698
 699
 700 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
 701 {
 702         struct net_device_context *ndevctx = netdev_priv(ndev);
 703         struct hv_device *hdev =  ndevctx->device_ctx;
 704         struct sockaddr *addr = p;
 705         char save_adr[ETH_ALEN];
 706         unsigned char save_aatype;
 707         int err;
 708
 709         memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
 710         save_aatype = ndev->addr_assign_type;
 711
 712         err = eth_mac_addr(ndev, p);
 713         if (err != 0)
 714                 return err;
 715
 716         err = rndis_filter_set_device_mac(hdev, addr->sa_data);
 717         if (err != 0) {
 718                 /* roll back to saved MAC */
 719                 memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
 720                 ndev->addr_assign_type = save_aatype;
 721         }
 722
 723         return err;
 724 }
 725
 726
 727 static const struct ethtool_ops ethtool_ops = {
 728         .get_drvinfo    = netvsc_get_drvinfo,
 729         .get_link       = ethtool_op_get_link,
 730 };
 731
 732 static const struct net_device_ops device_ops = {
 733         .ndo_open =                     netvsc_open,
 734         .ndo_stop =                     netvsc_close,
 735         .ndo_start_xmit =               netvsc_start_xmit,
 736         .ndo_set_rx_mode =              netvsc_set_multicast_list,
 737         .ndo_change_mtu =               netvsc_change_mtu,
 738         .ndo_validate_addr =            eth_validate_addr,
 739         .ndo_set_mac_address =          netvsc_set_mac_addr,
 740         .ndo_select_queue =             netvsc_select_queue,
 741 };
 742
 743 /*
 744  * Send GARP packet to network peers after migrations.
 745  * After Quick Migration, the network is not immediately operational in the
 746  * current context when receiving RNDIS_STATUS_MEDIA_CONNECT event. So, add
 747  * another netif_notify_peers() into a delayed work, otherwise GARP packet
 748  * will not be sent after quick migration, and cause network disconnection.
 749  * Also, we update the carrier status here.
 750  */
 751 static void netvsc_link_change(struct work_struct *w)
 752 {
 753         struct net_device_context *ndev_ctx;
 754         struct net_device *net;
 755         struct netvsc_device *net_device;
 756         struct rndis_device *rdev;
 757         bool notify;
 758
 759         rtnl_lock();
 760
 761         ndev_ctx = container_of(w, struct net_device_context, dwork.work);
 762         net_device = hv_get_drvdata(ndev_ctx->device_ctx);
 763         rdev = net_device->extension;
 764         net = net_device->ndev;
 765
 766         if (rdev->link_state) {
 767                 netif_carrier_off(net);
 768                 notify = false;
 769         } else {
 770                 netif_carrier_on(net);
 771                 notify = true;
 772         }
 773
 774         rtnl_unlock();
 775
 776         if (notify)
 777                 netdev_notify_peers(net);
 778 }
 779
 780
 781 static int netvsc_probe(struct hv_device *dev,
 782                         const struct hv_vmbus_device_id *dev_id)
 783 {
 784         struct net_device *net = NULL;
 785         struct net_device_context *net_device_ctx;
 786         struct netvsc_device_info device_info;
 787         struct netvsc_device *nvdev;
 788         int ret;
 789
 790         net = alloc_etherdev_mq(sizeof(struct net_device_context),
 791                                 num_online_cpus());
 792         if (!net)
 793                 return -ENOMEM;
 794
 795         netif_carrier_off(net);
 796
 797         net_device_ctx = netdev_priv(net);
 798         net_device_ctx->device_ctx = dev;
 799         hv_set_drvdata(dev, net);
 800         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
 801         INIT_WORK(&net_device_ctx->work, do_set_multicast);
 802
 803         net->netdev_ops = &device_ops;
 804
 805         net->hw_features = NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_IP_CSUM |
 806                                 NETIF_F_TSO;
 807         net->features = NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_SG | NETIF_F_RXCSUM |
 808                         NETIF_F_IP_CSUM | NETIF_F_TSO;
 809
 810         SET_ETHTOOL_OPS(net, &ethtool_ops);
 811         SET_NETDEV_DEV(net, &dev->device);
 812
 813         /* Notify the netvsc driver of the new device */
 814         device_info.ring_size = ring_size;
 815         ret = rndis_filter_device_add(dev, &device_info);
 816         if (ret != 0) {
 817                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
 818                 free_netdev(net);
 819                 hv_set_drvdata(dev, NULL);
 820                 return ret;
 821         }
 822         memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
 823
 824         nvdev = hv_get_drvdata(dev);
 825         netif_set_real_num_tx_queues(net, nvdev->num_chn);
 826         netif_set_real_num_rx_queues(net, nvdev->num_chn);
 827         dev_info(&dev->device, "real num tx,rx queues:%u, %u\n",
 828                  net->real_num_tx_queues, net->real_num_rx_queues);
 829
 830         ret = register_netdev(net);
 831         if (ret != 0) {
 832                 pr_err("Unable to register netdev.\n");
 833                 rndis_filter_device_remove(dev);
 834                 free_netdev(net);
 835         } else {
 836                 schedule_delayed_work(&net_device_ctx->dwork, 0);
 837         }
 838
 839         return ret;
 840 }
 841
 842 static int netvsc_remove(struct hv_device *dev)
 843 {
 844         struct net_device *net;
 845         struct net_device_context *ndev_ctx;
 846         struct netvsc_device *net_device;
 847
 848         net_device = hv_get_drvdata(dev);
 849         net = net_device->ndev;
 850
 851         if (net == NULL) {
 852                 dev_err(&dev->device, "No net device to remove\n");
 853                 return 0;
 854         }
 855
 856         net_device->start_remove = true;
 857
 858         ndev_ctx = netdev_priv(net);
 859         cancel_delayed_work_sync(&ndev_ctx->dwork);
 860         cancel_work_sync(&ndev_ctx->work);
 861
 862         /* Stop outbound asap */
 863         netif_tx_disable(net);
 864
 865         unregister_netdev(net);
 866
 867         /*
 868          * Call to the vsc driver to let it know that the device is being
 869          * removed
 870          */
 871         rndis_filter_device_remove(dev);
 872
 873         free_netdev(net);
 874         return 0;
 875 }
 876
 877 static const struct hv_vmbus_device_id id_table[] = {
 878         /* Network guid */
 879         { HV_NIC_GUID, },
 880         { },
 881 };
 882
 883 MODULE_DEVICE_TABLE(vmbus, id_table);
 884
 885 /* The one and only one */
 886 static struct  hv_driver netvsc_drv = {
 887         .name = KBUILD_MODNAME,
 888         .id_table = id_table,
 889         .probe = netvsc_probe,
 890         .remove = netvsc_remove,
 891 };
 892
 893 static void __exit netvsc_drv_exit(void)
 894 {
 895         vmbus_driver_unregister(&netvsc_drv);
 896 }
 897
 898 static int __init netvsc_drv_init(void)
 899 {
 900         if (ring_size < RING_SIZE_MIN) {
 901                 ring_size = RING_SIZE_MIN;
 902                 pr_info("Increased ring_size to %d (min allowed)\n",
 903                         ring_size);
 904         }
 905         return vmbus_driver_register(&netvsc_drv);
 906 }
 907
 908 MODULE_LICENSE("GPL");
 909 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
 910
 911 module_init(netvsc_drv_init);
 912 module_exit(netvsc_drv_exit);