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, void *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 = ((u8 *)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 flow_keys flow;
 194         int data_len;
 195
 196         if (!skb_flow_dissect(skb, &flow) ||
 197             !(flow.n_proto == htons(ETH_P_IP) ||
 198               flow.n_proto == htons(ETH_P_IPV6)))
 199                 return false;
 200
 201         if (flow.ip_proto == IPPROTO_TCP)
 202                 data_len = 12;
 203         else
 204                 data_len = 8;
 205
 206         *hash = comp_hash(netvsc_hash_key, HASH_KEYLEN, &flow, data_len);
 207
 208         return true;
 209 }
 210
 211 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
 212                         void *accel_priv, select_queue_fallback_t fallback)
 213 {
 214         struct net_device_context *net_device_ctx = netdev_priv(ndev);
 215         struct hv_device *hdev =  net_device_ctx->device_ctx;
 216         struct netvsc_device *nvsc_dev = hv_get_drvdata(hdev);
 217         u32 hash;
 218         u16 q_idx = 0;
 219
 220         if (nvsc_dev == NULL || ndev->real_num_tx_queues <= 1)
 221                 return 0;
 222
 223         if (netvsc_set_hash(&hash, skb)) {
 224                 q_idx = nvsc_dev->send_table[hash % VRSS_SEND_TAB_SIZE] %
 225                         ndev->real_num_tx_queues;
 226                 skb_set_hash(skb, hash, PKT_HASH_TYPE_L3);
 227         }
 228
 229         return q_idx;
 230 }
 231
 232 static void netvsc_xmit_completion(void *context)
 233 {
 234         struct hv_netvsc_packet *packet = (struct hv_netvsc_packet *)context;
 235         struct sk_buff *skb = (struct sk_buff *)
 236                 (unsigned long)packet->send_completion_tid;
 237         u32 index = packet->send_buf_index;
 238
 239         kfree(packet);
 240
 241         if (skb && (index == NETVSC_INVALID_INDEX))
 242                 dev_kfree_skb_any(skb);
 243 }
 244
 245 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
 246                         struct hv_page_buffer *pb)
 247 {
 248         int j = 0;
 249
 250         /* Deal with compund pages by ignoring unused part
 251          * of the page.
 252          */
 253         page += (offset >> PAGE_SHIFT);
 254         offset &= ~PAGE_MASK;
 255
 256         while (len > 0) {
 257                 unsigned long bytes;
 258
 259                 bytes = PAGE_SIZE - offset;
 260                 if (bytes > len)
 261                         bytes = len;
 262                 pb[j].pfn = page_to_pfn(page);
 263                 pb[j].offset = offset;
 264                 pb[j].len = bytes;
 265
 266                 offset += bytes;
 267                 len -= bytes;
 268
 269                 if (offset == PAGE_SIZE && len) {
 270                         page++;
 271                         offset = 0;
 272                         j++;
 273                 }
 274         }
 275
 276         return j + 1;
 277 }
 278
 279 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
 280                            struct hv_page_buffer *pb)
 281 {
 282         u32 slots_used = 0;
 283         char *data = skb->data;
 284         int frags = skb_shinfo(skb)->nr_frags;
 285         int i;
 286
 287         /* The packet is laid out thus:
 288          * 1. hdr
 289          * 2. skb linear data
 290          * 3. skb fragment data
 291          */
 292         if (hdr != NULL)
 293                 slots_used += fill_pg_buf(virt_to_page(hdr),
 294                                         offset_in_page(hdr),
 295                                         len, &pb[slots_used]);
 296
 297         slots_used += fill_pg_buf(virt_to_page(data),
 298                                 offset_in_page(data),
 299                                 skb_headlen(skb), &pb[slots_used]);
 300
 301         for (i = 0; i < frags; i++) {
 302                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 303
 304                 slots_used += fill_pg_buf(skb_frag_page(frag),
 305                                         frag->page_offset,
 306                                         skb_frag_size(frag), &pb[slots_used]);
 307         }
 308         return slots_used;
 309 }
 310
 311 static int count_skb_frag_slots(struct sk_buff *skb)
 312 {
 313         int i, frags = skb_shinfo(skb)->nr_frags;
 314         int pages = 0;
 315
 316         for (i = 0; i < frags; i++) {
 317                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 318                 unsigned long size = skb_frag_size(frag);
 319                 unsigned long offset = frag->page_offset;
 320
 321                 /* Skip unused frames from start of page */
 322                 offset &= ~PAGE_MASK;
 323                 pages += PFN_UP(offset + size);
 324         }
 325         return pages;
 326 }
 327
 328 static int netvsc_get_slots(struct sk_buff *skb)
 329 {
 330         char *data = skb->data;
 331         unsigned int offset = offset_in_page(data);
 332         unsigned int len = skb_headlen(skb);
 333         int slots;
 334         int frag_slots;
 335
 336         slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
 337         frag_slots = count_skb_frag_slots(skb);
 338         return slots + frag_slots;
 339 }
 340
 341 static u32 get_net_transport_info(struct sk_buff *skb, u32 *trans_off)
 342 {
 343         u32 ret_val = TRANSPORT_INFO_NOT_IP;
 344
 345         if ((eth_hdr(skb)->h_proto != htons(ETH_P_IP)) &&
 346                 (eth_hdr(skb)->h_proto != htons(ETH_P_IPV6))) {
 347                 goto not_ip;
 348         }
 349
 350         *trans_off = skb_transport_offset(skb);
 351
 352         if ((eth_hdr(skb)->h_proto == htons(ETH_P_IP))) {
 353                 struct iphdr *iphdr = ip_hdr(skb);
 354
 355                 if (iphdr->protocol == IPPROTO_TCP)
 356                         ret_val = TRANSPORT_INFO_IPV4_TCP;
 357                 else if (iphdr->protocol == IPPROTO_UDP)
 358                         ret_val = TRANSPORT_INFO_IPV4_UDP;
 359         } else {
 360                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
 361                         ret_val = TRANSPORT_INFO_IPV6_TCP;
 362                 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
 363                         ret_val = TRANSPORT_INFO_IPV6_UDP;
 364         }
 365
 366 not_ip:
 367         return ret_val;
 368 }
 369
 370 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
 371 {
 372         struct net_device_context *net_device_ctx = netdev_priv(net);
 373         struct hv_netvsc_packet *packet;
 374         int ret;
 375         unsigned int num_data_pgs;
 376         struct rndis_message *rndis_msg;
 377         struct rndis_packet *rndis_pkt;
 378         u32 rndis_msg_size;
 379         bool isvlan;
 380         struct rndis_per_packet_info *ppi;
 381         struct ndis_tcp_ip_checksum_info *csum_info;
 382         struct ndis_tcp_lso_info *lso_info;
 383         int  hdr_offset;
 384         u32 net_trans_info;
 385         u32 hash;
 386         u32 skb_length = skb->len;
 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 + NDIS_CSUM_PPI_SIZE +
 406                          NDIS_LSO_PPI_SIZE + NDIS_HASH_PPI_SIZE, GFP_ATOMIC);
 407         if (!packet) {
 408                 /* out of memory, drop packet */
 409                 netdev_err(net, "unable to allocate hv_netvsc_packet\n");
 410
 411                 dev_kfree_skb(skb);
 412                 net->stats.tx_dropped++;
 413                 return NETDEV_TX_OK;
 414         }
 415
 416         packet->vlan_tci = skb->vlan_tci;
 417
 418         packet->q_idx = skb_get_queue_mapping(skb);
 419
 420         packet->is_data_pkt = true;
 421         packet->total_data_buflen = skb->len;
 422
 423         packet->rndis_msg = (struct rndis_message *)((unsigned long)packet +
 424                                 sizeof(struct hv_netvsc_packet) +
 425                                 (num_data_pgs * sizeof(struct hv_page_buffer)));
 426
 427         /* Set the completion routine */
 428         packet->send_completion = netvsc_xmit_completion;
 429         packet->send_completion_ctx = packet;
 430         packet->send_completion_tid = (unsigned long)skb;
 431
 432         isvlan = packet->vlan_tci & VLAN_TAG_PRESENT;
 433
 434         /* Add the rndis header */
 435         rndis_msg = packet->rndis_msg;
 436         rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
 437         rndis_msg->msg_len = packet->total_data_buflen;
 438         rndis_pkt = &rndis_msg->msg.pkt;
 439         rndis_pkt->data_offset = sizeof(struct rndis_packet);
 440         rndis_pkt->data_len = packet->total_data_buflen;
 441         rndis_pkt->per_pkt_info_offset = sizeof(struct rndis_packet);
 442
 443         rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
 444
 445         hash = skb_get_hash_raw(skb);
 446         if (hash != 0 && net->real_num_tx_queues > 1) {
 447                 rndis_msg_size += NDIS_HASH_PPI_SIZE;
 448                 ppi = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
 449                                     NBL_HASH_VALUE);
 450                 *(u32 *)((void *)ppi + ppi->ppi_offset) = hash;
 451         }
 452
 453         if (isvlan) {
 454                 struct ndis_pkt_8021q_info *vlan;
 455
 456                 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
 457                 ppi = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
 458                                         IEEE_8021Q_INFO);
 459                 vlan = (struct ndis_pkt_8021q_info *)((void *)ppi +
 460                                                 ppi->ppi_offset);
 461                 vlan->vlanid = packet->vlan_tci & VLAN_VID_MASK;
 462                 vlan->pri = (packet->vlan_tci & VLAN_PRIO_MASK) >>
 463                                 VLAN_PRIO_SHIFT;
 464         }
 465
 466         net_trans_info = get_net_transport_info(skb, &hdr_offset);
 467         if (net_trans_info == TRANSPORT_INFO_NOT_IP)
 468                 goto do_send;
 469
 470         /*
 471          * Setup the sendside checksum offload only if this is not a
 472          * GSO packet.
 473          */
 474         if (skb_is_gso(skb))
 475                 goto do_lso;
 476
 477         if ((skb->ip_summed == CHECKSUM_NONE) ||
 478             (skb->ip_summed == CHECKSUM_UNNECESSARY))
 479                 goto do_send;
 480
 481         rndis_msg_size += NDIS_CSUM_PPI_SIZE;
 482         ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
 483                             TCPIP_CHKSUM_PKTINFO);
 484
 485         csum_info = (struct ndis_tcp_ip_checksum_info *)((void *)ppi +
 486                         ppi->ppi_offset);
 487
 488         if (net_trans_info & (INFO_IPV4 << 16))
 489                 csum_info->transmit.is_ipv4 = 1;
 490         else
 491                 csum_info->transmit.is_ipv6 = 1;
 492
 493         if (net_trans_info & INFO_TCP) {
 494                 csum_info->transmit.tcp_checksum = 1;
 495                 csum_info->transmit.tcp_header_offset = hdr_offset;
 496         } else if (net_trans_info & INFO_UDP) {
 497                 /* UDP checksum offload is not supported on ws2008r2.
 498                  * Furthermore, on ws2012 and ws2012r2, there are some
 499                  * issues with udp checksum offload from Linux guests.
 500                  * (these are host issues).
 501                  * For now compute the checksum here.
 502                  */
 503                 struct udphdr *uh;
 504                 u16 udp_len;
 505
 506                 ret = skb_cow_head(skb, 0);
 507                 if (ret)
 508                         goto drop;
 509
 510                 uh = udp_hdr(skb);
 511                 udp_len = ntohs(uh->len);
 512                 uh->check = 0;
 513                 uh->check = csum_tcpudp_magic(ip_hdr(skb)->saddr,
 514                                               ip_hdr(skb)->daddr,
 515                                               udp_len, IPPROTO_UDP,
 516                                               csum_partial(uh, udp_len, 0));
 517                 if (uh->check == 0)
 518                         uh->check = CSUM_MANGLED_0;
 519
 520                 csum_info->transmit.udp_checksum = 0;
 521         }
 522         goto do_send;
 523
 524 do_lso:
 525         rndis_msg_size += NDIS_LSO_PPI_SIZE;
 526         ppi = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
 527                             TCP_LARGESEND_PKTINFO);
 528
 529         lso_info = (struct ndis_tcp_lso_info *)((void *)ppi +
 530                         ppi->ppi_offset);
 531
 532         lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
 533         if (net_trans_info & (INFO_IPV4 << 16)) {
 534                 lso_info->lso_v2_transmit.ip_version =
 535                         NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
 536                 ip_hdr(skb)->tot_len = 0;
 537                 ip_hdr(skb)->check = 0;
 538                 tcp_hdr(skb)->check =
 539                 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
 540                                    ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 541         } else {
 542                 lso_info->lso_v2_transmit.ip_version =
 543                         NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
 544                 ipv6_hdr(skb)->payload_len = 0;
 545                 tcp_hdr(skb)->check =
 546                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 547                                 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 548         }
 549         lso_info->lso_v2_transmit.tcp_header_offset = hdr_offset;
 550         lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
 551
 552 do_send:
 553         /* Start filling in the page buffers with the rndis hdr */
 554         rndis_msg->msg_len += rndis_msg_size;
 555         packet->total_data_buflen = rndis_msg->msg_len;
 556         packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
 557                                         skb, &packet->page_buf[0]);
 558
 559         ret = netvsc_send(net_device_ctx->device_ctx, packet);
 560
 561 drop:
 562         if (ret == 0) {
 563                 net->stats.tx_bytes += skb_length;
 564                 net->stats.tx_packets++;
 565         } else {
 566                 kfree(packet);
 567                 if (ret != -EAGAIN) {
 568                         dev_kfree_skb_any(skb);
 569                         net->stats.tx_dropped++;
 570                 }
 571         }
 572
 573         return (ret == -EAGAIN) ? NETDEV_TX_BUSY : NETDEV_TX_OK;
 574 }
 575
 576 /*
 577  * netvsc_linkstatus_callback - Link up/down notification
 578  */
 579 void netvsc_linkstatus_callback(struct hv_device *device_obj,
 580                                 struct rndis_message *resp)
 581 {
 582         struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
 583         struct net_device *net;
 584         struct net_device_context *ndev_ctx;
 585         struct netvsc_device *net_device;
 586         struct rndis_device *rdev;
 587
 588         net_device = hv_get_drvdata(device_obj);
 589         rdev = net_device->extension;
 590
 591         switch (indicate->status) {
 592         case RNDIS_STATUS_MEDIA_CONNECT:
 593                 rdev->link_state = false;
 594                 break;
 595         case RNDIS_STATUS_MEDIA_DISCONNECT:
 596                 rdev->link_state = true;
 597                 break;
 598         case RNDIS_STATUS_NETWORK_CHANGE:
 599                 rdev->link_change = true;
 600                 break;
 601         default:
 602                 return;
 603         }
 604
 605         net = net_device->ndev;
 606
 607         if (!net || net->reg_state != NETREG_REGISTERED)
 608                 return;
 609
 610         ndev_ctx = netdev_priv(net);
 611         if (!rdev->link_state) {
 612                 schedule_delayed_work(&ndev_ctx->dwork, 0);
 613                 schedule_delayed_work(&ndev_ctx->dwork, msecs_to_jiffies(20));
 614         } else {
 615                 schedule_delayed_work(&ndev_ctx->dwork, 0);
 616         }
 617 }
 618
 619 /*
 620  * netvsc_recv_callback -  Callback when we receive a packet from the
 621  * "wire" on the specified device.
 622  */
 623 int netvsc_recv_callback(struct hv_device *device_obj,
 624                                 struct hv_netvsc_packet *packet,
 625                                 struct ndis_tcp_ip_checksum_info *csum_info)
 626 {
 627         struct net_device *net;
 628         struct sk_buff *skb;
 629
 630         net = ((struct netvsc_device *)hv_get_drvdata(device_obj))->ndev;
 631         if (!net || net->reg_state != NETREG_REGISTERED) {
 632                 packet->status = NVSP_STAT_FAIL;
 633                 return 0;
 634         }
 635
 636         /* Allocate a skb - TODO direct I/O to pages? */
 637         skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
 638         if (unlikely(!skb)) {
 639                 ++net->stats.rx_dropped;
 640                 packet->status = NVSP_STAT_FAIL;
 641                 return 0;
 642         }
 643
 644         /*
 645          * Copy to skb. This copy is needed here since the memory pointed by
 646          * hv_netvsc_packet cannot be deallocated
 647          */
 648         memcpy(skb_put(skb, packet->total_data_buflen), packet->data,
 649                 packet->total_data_buflen);
 650
 651         skb->protocol = eth_type_trans(skb, net);
 652         if (csum_info) {
 653                 /* We only look at the IP checksum here.
 654                  * Should we be dropping the packet if checksum
 655                  * failed? How do we deal with other checksums - TCP/UDP?
 656                  */
 657                 if (csum_info->receive.ip_checksum_succeeded)
 658                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 659                 else
 660                         skb->ip_summed = CHECKSUM_NONE;
 661         }
 662
 663         if (packet->vlan_tci & VLAN_TAG_PRESENT)
 664                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 665                                        packet->vlan_tci);
 666
 667         skb_record_rx_queue(skb, packet->channel->
 668                             offermsg.offer.sub_channel_index);
 669
 670         net->stats.rx_packets++;
 671         net->stats.rx_bytes += packet->total_data_buflen;
 672
 673         /*
 674          * Pass the skb back up. Network stack will deallocate the skb when it
 675          * is done.
 676          * TODO - use NAPI?
 677          */
 678         netif_rx(skb);
 679
 680         return 0;
 681 }
 682
 683 static void netvsc_get_drvinfo(struct net_device *net,
 684                                struct ethtool_drvinfo *info)
 685 {
 686         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
 687         strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
 688 }
 689
 690 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
 691 {
 692         struct net_device_context *ndevctx = netdev_priv(ndev);
 693         struct hv_device *hdev =  ndevctx->device_ctx;
 694         struct netvsc_device *nvdev = hv_get_drvdata(hdev);
 695         struct netvsc_device_info device_info;
 696         int limit = ETH_DATA_LEN;
 697
 698         if (nvdev == NULL || nvdev->destroy)
 699                 return -ENODEV;
 700
 701         if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
 702                 limit = NETVSC_MTU - ETH_HLEN;
 703
 704         /* Hyper-V hosts don't support MTU < ETH_DATA_LEN (1500) */
 705         if (mtu < ETH_DATA_LEN || mtu > limit)
 706                 return -EINVAL;
 707
 708         nvdev->start_remove = true;
 709         cancel_work_sync(&ndevctx->work);
 710         netif_tx_disable(ndev);
 711         rndis_filter_device_remove(hdev);
 712
 713         ndev->mtu = mtu;
 714
 715         ndevctx->device_ctx = hdev;
 716         hv_set_drvdata(hdev, ndev);
 717         device_info.ring_size = ring_size;
 718         rndis_filter_device_add(hdev, &device_info);
 719         netif_tx_wake_all_queues(ndev);
 720
 721         return 0;
 722 }
 723
 724
 725 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
 726 {
 727         struct net_device_context *ndevctx = netdev_priv(ndev);
 728         struct hv_device *hdev =  ndevctx->device_ctx;
 729         struct sockaddr *addr = p;
 730         char save_adr[ETH_ALEN];
 731         unsigned char save_aatype;
 732         int err;
 733
 734         memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
 735         save_aatype = ndev->addr_assign_type;
 736
 737         err = eth_mac_addr(ndev, p);
 738         if (err != 0)
 739                 return err;
 740
 741         err = rndis_filter_set_device_mac(hdev, addr->sa_data);
 742         if (err != 0) {
 743                 /* roll back to saved MAC */
 744                 memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
 745                 ndev->addr_assign_type = save_aatype;
 746         }
 747
 748         return err;
 749 }
 750
 751 #ifdef CONFIG_NET_POLL_CONTROLLER
 752 static void netvsc_poll_controller(struct net_device *net)
 753 {
 754         /* As netvsc_start_xmit() works synchronous we don't have to
 755          * trigger anything here.
 756          */
 757 }
 758 #endif
 759
 760 static const struct ethtool_ops ethtool_ops = {
 761         .get_drvinfo    = netvsc_get_drvinfo,
 762         .get_link       = ethtool_op_get_link,
 763 };
 764
 765 static const struct net_device_ops device_ops = {
 766         .ndo_open =                     netvsc_open,
 767         .ndo_stop =                     netvsc_close,
 768         .ndo_start_xmit =               netvsc_start_xmit,
 769         .ndo_set_rx_mode =              netvsc_set_multicast_list,
 770         .ndo_change_mtu =               netvsc_change_mtu,
 771         .ndo_validate_addr =            eth_validate_addr,
 772         .ndo_set_mac_address =          netvsc_set_mac_addr,
 773         .ndo_select_queue =             netvsc_select_queue,
 774 #ifdef CONFIG_NET_POLL_CONTROLLER
 775         .ndo_poll_controller =          netvsc_poll_controller,
 776 #endif
 777 };
 778
 779 /*
 780  * Send GARP packet to network peers after migrations.
 781  * After Quick Migration, the network is not immediately operational in the
 782  * current context when receiving RNDIS_STATUS_MEDIA_CONNECT event. So, add
 783  * another netif_notify_peers() into a delayed work, otherwise GARP packet
 784  * will not be sent after quick migration, and cause network disconnection.
 785  * Also, we update the carrier status here.
 786  */
 787 static void netvsc_link_change(struct work_struct *w)
 788 {
 789         struct net_device_context *ndev_ctx;
 790         struct net_device *net;
 791         struct netvsc_device *net_device;
 792         struct rndis_device *rdev;
 793         bool notify, refresh = false;
 794         char *argv[] = { "/etc/init.d/network", "restart", NULL };
 795         char *envp[] = { "HOME=/", "PATH=/sbin:/usr/sbin:/bin:/usr/bin", NULL };
 796
 797         rtnl_lock();
 798
 799         ndev_ctx = container_of(w, struct net_device_context, dwork.work);
 800         net_device = hv_get_drvdata(ndev_ctx->device_ctx);
 801         rdev = net_device->extension;
 802         net = net_device->ndev;
 803
 804         if (rdev->link_state) {
 805                 netif_carrier_off(net);
 806                 notify = false;
 807         } else {
 808                 netif_carrier_on(net);
 809                 notify = true;
 810                 if (rdev->link_change) {
 811                         rdev->link_change = false;
 812                         refresh = true;
 813                 }
 814         }
 815
 816         rtnl_unlock();
 817
 818         if (refresh)
 819                 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
 820
 821         if (notify)
 822                 netdev_notify_peers(net);
 823 }
 824
 825
 826 static int netvsc_probe(struct hv_device *dev,
 827                         const struct hv_vmbus_device_id *dev_id)
 828 {
 829         struct net_device *net = NULL;
 830         struct net_device_context *net_device_ctx;
 831         struct netvsc_device_info device_info;
 832         struct netvsc_device *nvdev;
 833         int ret;
 834
 835         net = alloc_etherdev_mq(sizeof(struct net_device_context),
 836                                 num_online_cpus());
 837         if (!net)
 838                 return -ENOMEM;
 839
 840         netif_carrier_off(net);
 841
 842         net_device_ctx = netdev_priv(net);
 843         net_device_ctx->device_ctx = dev;
 844         hv_set_drvdata(dev, net);
 845         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
 846         INIT_WORK(&net_device_ctx->work, do_set_multicast);
 847
 848         net->netdev_ops = &device_ops;
 849
 850         net->hw_features = NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_IP_CSUM |
 851                                 NETIF_F_TSO;
 852         net->features = NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_SG | NETIF_F_RXCSUM |
 853                         NETIF_F_IP_CSUM | NETIF_F_TSO;
 854
 855         net->ethtool_ops = &ethtool_ops;
 856         SET_NETDEV_DEV(net, &dev->device);
 857
 858         /* Notify the netvsc driver of the new device */
 859         device_info.ring_size = ring_size;
 860         ret = rndis_filter_device_add(dev, &device_info);
 861         if (ret != 0) {
 862                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
 863                 free_netdev(net);
 864                 hv_set_drvdata(dev, NULL);
 865                 return ret;
 866         }
 867         memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
 868
 869         nvdev = hv_get_drvdata(dev);
 870         netif_set_real_num_tx_queues(net, nvdev->num_chn);
 871         netif_set_real_num_rx_queues(net, nvdev->num_chn);
 872
 873         ret = register_netdev(net);
 874         if (ret != 0) {
 875                 pr_err("Unable to register netdev.\n");
 876                 rndis_filter_device_remove(dev);
 877                 free_netdev(net);
 878         } else {
 879                 schedule_delayed_work(&net_device_ctx->dwork, 0);
 880         }
 881
 882         return ret;
 883 }
 884
 885 static int netvsc_remove(struct hv_device *dev)
 886 {
 887         struct net_device *net;
 888         struct net_device_context *ndev_ctx;
 889         struct netvsc_device *net_device;
 890
 891         net_device = hv_get_drvdata(dev);
 892         net = net_device->ndev;
 893
 894         if (net == NULL) {
 895                 dev_err(&dev->device, "No net device to remove\n");
 896                 return 0;
 897         }
 898
 899         net_device->start_remove = true;
 900
 901         ndev_ctx = netdev_priv(net);
 902         cancel_delayed_work_sync(&ndev_ctx->dwork);
 903         cancel_work_sync(&ndev_ctx->work);
 904
 905         /* Stop outbound asap */
 906         netif_tx_disable(net);
 907
 908         unregister_netdev(net);
 909
 910         /*
 911          * Call to the vsc driver to let it know that the device is being
 912          * removed
 913          */
 914         rndis_filter_device_remove(dev);
 915
 916         free_netdev(net);
 917         return 0;
 918 }
 919
 920 static const struct hv_vmbus_device_id id_table[] = {
 921         /* Network guid */
 922         { HV_NIC_GUID, },
 923         { },
 924 };
 925
 926 MODULE_DEVICE_TABLE(vmbus, id_table);
 927
 928 /* The one and only one */
 929 static struct  hv_driver netvsc_drv = {
 930         .name = KBUILD_MODNAME,
 931         .id_table = id_table,
 932         .probe = netvsc_probe,
 933         .remove = netvsc_remove,
 934 };
 935
 936 static void __exit netvsc_drv_exit(void)
 937 {
 938         vmbus_driver_unregister(&netvsc_drv);
 939 }
 940
 941 static int __init netvsc_drv_init(void)
 942 {
 943         if (ring_size < RING_SIZE_MIN) {
 944                 ring_size = RING_SIZE_MIN;
 945                 pr_info("Increased ring_size to %d (min allowed)\n",
 946                         ring_size);
 947         }
 948         return vmbus_driver_register(&netvsc_drv);
 949 }
 950
 951 MODULE_LICENSE("GPL");
 952 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
 953
 954 module_init(netvsc_drv_init);
 955 module_exit(netvsc_drv_exit);