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) || flow.n_proto != htons(ETH_P_IP))
 197                 return false;
 198
 199         if (flow.ip_proto == IPPROTO_TCP)
 200                 data_len = 12;
 201         else
 202                 data_len = 8;
 203
 204         *hash = comp_hash(netvsc_hash_key, HASH_KEYLEN, &flow, data_len);
 205
 206         return true;
 207 }
 208
 209 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
 210                         void *accel_priv, select_queue_fallback_t fallback)
 211 {
 212         struct net_device_context *net_device_ctx = netdev_priv(ndev);
 213         struct hv_device *hdev =  net_device_ctx->device_ctx;
 214         struct netvsc_device *nvsc_dev = hv_get_drvdata(hdev);
 215         u32 hash;
 216         u16 q_idx = 0;
 217
 218         if (nvsc_dev == NULL || ndev->real_num_tx_queues <= 1)
 219                 return 0;
 220
 221         if (netvsc_set_hash(&hash, skb)) {
 222                 q_idx = nvsc_dev->send_table[hash % VRSS_SEND_TAB_SIZE] %
 223                         ndev->real_num_tx_queues;
 224                 skb_set_hash(skb, hash, PKT_HASH_TYPE_L3);
 225         }
 226
 227         return q_idx;
 228 }
 229
 230 static void netvsc_xmit_completion(void *context)
 231 {
 232         struct hv_netvsc_packet *packet = (struct hv_netvsc_packet *)context;
 233         struct sk_buff *skb = (struct sk_buff *)
 234                 (unsigned long)packet->send_completion_tid;
 235         u32 index = packet->send_buf_index;
 236
 237         kfree(packet);
 238
 239         if (skb && (index == NETVSC_INVALID_INDEX))
 240                 dev_kfree_skb_any(skb);
 241 }
 242
 243 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
 244                         struct hv_page_buffer *pb)
 245 {
 246         int j = 0;
 247
 248         /* Deal with compund pages by ignoring unused part
 249          * of the page.
 250          */
 251         page += (offset >> PAGE_SHIFT);
 252         offset &= ~PAGE_MASK;
 253
 254         while (len > 0) {
 255                 unsigned long bytes;
 256
 257                 bytes = PAGE_SIZE - offset;
 258                 if (bytes > len)
 259                         bytes = len;
 260                 pb[j].pfn = page_to_pfn(page);
 261                 pb[j].offset = offset;
 262                 pb[j].len = bytes;
 263
 264                 offset += bytes;
 265                 len -= bytes;
 266
 267                 if (offset == PAGE_SIZE && len) {
 268                         page++;
 269                         offset = 0;
 270                         j++;
 271                 }
 272         }
 273
 274         return j + 1;
 275 }
 276
 277 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
 278                            struct hv_page_buffer *pb)
 279 {
 280         u32 slots_used = 0;
 281         char *data = skb->data;
 282         int frags = skb_shinfo(skb)->nr_frags;
 283         int i;
 284
 285         /* The packet is laid out thus:
 286          * 1. hdr
 287          * 2. skb linear data
 288          * 3. skb fragment data
 289          */
 290         if (hdr != NULL)
 291                 slots_used += fill_pg_buf(virt_to_page(hdr),
 292                                         offset_in_page(hdr),
 293                                         len, &pb[slots_used]);
 294
 295         slots_used += fill_pg_buf(virt_to_page(data),
 296                                 offset_in_page(data),
 297                                 skb_headlen(skb), &pb[slots_used]);
 298
 299         for (i = 0; i < frags; i++) {
 300                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 301
 302                 slots_used += fill_pg_buf(skb_frag_page(frag),
 303                                         frag->page_offset,
 304                                         skb_frag_size(frag), &pb[slots_used]);
 305         }
 306         return slots_used;
 307 }
 308
 309 static int count_skb_frag_slots(struct sk_buff *skb)
 310 {
 311         int i, frags = skb_shinfo(skb)->nr_frags;
 312         int pages = 0;
 313
 314         for (i = 0; i < frags; i++) {
 315                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 316                 unsigned long size = skb_frag_size(frag);
 317                 unsigned long offset = frag->page_offset;
 318
 319                 /* Skip unused frames from start of page */
 320                 offset &= ~PAGE_MASK;
 321                 pages += PFN_UP(offset + size);
 322         }
 323         return pages;
 324 }
 325
 326 static int netvsc_get_slots(struct sk_buff *skb)
 327 {
 328         char *data = skb->data;
 329         unsigned int offset = offset_in_page(data);
 330         unsigned int len = skb_headlen(skb);
 331         int slots;
 332         int frag_slots;
 333
 334         slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
 335         frag_slots = count_skb_frag_slots(skb);
 336         return slots + frag_slots;
 337 }
 338
 339 static u32 get_net_transport_info(struct sk_buff *skb, u32 *trans_off)
 340 {
 341         u32 ret_val = TRANSPORT_INFO_NOT_IP;
 342
 343         if ((eth_hdr(skb)->h_proto != htons(ETH_P_IP)) &&
 344                 (eth_hdr(skb)->h_proto != htons(ETH_P_IPV6))) {
 345                 goto not_ip;
 346         }
 347
 348         *trans_off = skb_transport_offset(skb);
 349
 350         if ((eth_hdr(skb)->h_proto == htons(ETH_P_IP))) {
 351                 struct iphdr *iphdr = ip_hdr(skb);
 352
 353                 if (iphdr->protocol == IPPROTO_TCP)
 354                         ret_val = TRANSPORT_INFO_IPV4_TCP;
 355                 else if (iphdr->protocol == IPPROTO_UDP)
 356                         ret_val = TRANSPORT_INFO_IPV4_UDP;
 357         } else {
 358                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
 359                         ret_val = TRANSPORT_INFO_IPV6_TCP;
 360                 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
 361                         ret_val = TRANSPORT_INFO_IPV6_UDP;
 362         }
 363
 364 not_ip:
 365         return ret_val;
 366 }
 367
 368 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
 369 {
 370         struct net_device_context *net_device_ctx = netdev_priv(net);
 371         struct hv_netvsc_packet *packet;
 372         int ret;
 373         unsigned int num_data_pgs;
 374         struct rndis_message *rndis_msg;
 375         struct rndis_packet *rndis_pkt;
 376         u32 rndis_msg_size;
 377         bool isvlan;
 378         struct rndis_per_packet_info *ppi;
 379         struct ndis_tcp_ip_checksum_info *csum_info;
 380         struct ndis_tcp_lso_info *lso_info;
 381         int  hdr_offset;
 382         u32 net_trans_info;
 383         u32 hash;
 384         u32 skb_length = skb->len;
 385
 386
 387         /* We will atmost need two pages to describe the rndis
 388          * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
 389          * of pages in a single packet.
 390          */
 391         num_data_pgs = netvsc_get_slots(skb) + 2;
 392         if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
 393                 netdev_err(net, "Packet too big: %u\n", skb->len);
 394                 dev_kfree_skb(skb);
 395                 net->stats.tx_dropped++;
 396                 return NETDEV_TX_OK;
 397         }
 398
 399         /* Allocate a netvsc packet based on # of frags. */
 400         packet = kzalloc(sizeof(struct hv_netvsc_packet) +
 401                          (num_data_pgs * sizeof(struct hv_page_buffer)) +
 402                          sizeof(struct rndis_message) +
 403                          NDIS_VLAN_PPI_SIZE + NDIS_CSUM_PPI_SIZE +
 404                          NDIS_LSO_PPI_SIZE + NDIS_HASH_PPI_SIZE, GFP_ATOMIC);
 405         if (!packet) {
 406                 /* out of memory, drop packet */
 407                 netdev_err(net, "unable to allocate hv_netvsc_packet\n");
 408
 409                 dev_kfree_skb(skb);
 410                 net->stats.tx_dropped++;
 411                 return NETDEV_TX_OK;
 412         }
 413
 414         packet->vlan_tci = skb->vlan_tci;
 415
 416         packet->q_idx = skb_get_queue_mapping(skb);
 417
 418         packet->is_data_pkt = true;
 419         packet->total_data_buflen = skb->len;
 420
 421         packet->rndis_msg = (struct rndis_message *)((unsigned long)packet +
 422                                 sizeof(struct hv_netvsc_packet) +
 423                                 (num_data_pgs * sizeof(struct hv_page_buffer)));
 424
 425         /* Set the completion routine */
 426         packet->send_completion = netvsc_xmit_completion;
 427         packet->send_completion_ctx = packet;
 428         packet->send_completion_tid = (unsigned long)skb;
 429
 430         isvlan = packet->vlan_tci & VLAN_TAG_PRESENT;
 431
 432         /* Add the rndis header */
 433         rndis_msg = packet->rndis_msg;
 434         rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
 435         rndis_msg->msg_len = packet->total_data_buflen;
 436         rndis_pkt = &rndis_msg->msg.pkt;
 437         rndis_pkt->data_offset = sizeof(struct rndis_packet);
 438         rndis_pkt->data_len = packet->total_data_buflen;
 439         rndis_pkt->per_pkt_info_offset = sizeof(struct rndis_packet);
 440
 441         rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
 442
 443         hash = skb_get_hash_raw(skb);
 444         if (hash != 0 && net->real_num_tx_queues > 1) {
 445                 rndis_msg_size += NDIS_HASH_PPI_SIZE;
 446                 ppi = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
 447                                     NBL_HASH_VALUE);
 448                 *(u32 *)((void *)ppi + ppi->ppi_offset) = hash;
 449         }
 450
 451         if (isvlan) {
 452                 struct ndis_pkt_8021q_info *vlan;
 453
 454                 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
 455                 ppi = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
 456                                         IEEE_8021Q_INFO);
 457                 vlan = (struct ndis_pkt_8021q_info *)((void *)ppi +
 458                                                 ppi->ppi_offset);
 459                 vlan->vlanid = packet->vlan_tci & VLAN_VID_MASK;
 460                 vlan->pri = (packet->vlan_tci & VLAN_PRIO_MASK) >>
 461                                 VLAN_PRIO_SHIFT;
 462         }
 463
 464         net_trans_info = get_net_transport_info(skb, &hdr_offset);
 465         if (net_trans_info == TRANSPORT_INFO_NOT_IP)
 466                 goto do_send;
 467
 468         /*
 469          * Setup the sendside checksum offload only if this is not a
 470          * GSO packet.
 471          */
 472         if (skb_is_gso(skb))
 473                 goto do_lso;
 474
 475         if ((skb->ip_summed == CHECKSUM_NONE) ||
 476             (skb->ip_summed == CHECKSUM_UNNECESSARY))
 477                 goto do_send;
 478
 479         rndis_msg_size += NDIS_CSUM_PPI_SIZE;
 480         ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
 481                             TCPIP_CHKSUM_PKTINFO);
 482
 483         csum_info = (struct ndis_tcp_ip_checksum_info *)((void *)ppi +
 484                         ppi->ppi_offset);
 485
 486         if (net_trans_info & (INFO_IPV4 << 16))
 487                 csum_info->transmit.is_ipv4 = 1;
 488         else
 489                 csum_info->transmit.is_ipv6 = 1;
 490
 491         if (net_trans_info & INFO_TCP) {
 492                 csum_info->transmit.tcp_checksum = 1;
 493                 csum_info->transmit.tcp_header_offset = hdr_offset;
 494         } else if (net_trans_info & INFO_UDP) {
 495                 /* UDP checksum offload is not supported on ws2008r2.
 496                  * Furthermore, on ws2012 and ws2012r2, there are some
 497                  * issues with udp checksum offload from Linux guests.
 498                  * (these are host issues).
 499                  * For now compute the checksum here.
 500                  */
 501                 struct udphdr *uh;
 502                 u16 udp_len;
 503
 504                 ret = skb_cow_head(skb, 0);
 505                 if (ret)
 506                         goto drop;
 507
 508                 uh = udp_hdr(skb);
 509                 udp_len = ntohs(uh->len);
 510                 uh->check = 0;
 511                 uh->check = csum_tcpudp_magic(ip_hdr(skb)->saddr,
 512                                               ip_hdr(skb)->daddr,
 513                                               udp_len, IPPROTO_UDP,
 514                                               csum_partial(uh, udp_len, 0));
 515                 if (uh->check == 0)
 516                         uh->check = CSUM_MANGLED_0;
 517
 518                 csum_info->transmit.udp_checksum = 0;
 519         }
 520         goto do_send;
 521
 522 do_lso:
 523         rndis_msg_size += NDIS_LSO_PPI_SIZE;
 524         ppi = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
 525                             TCP_LARGESEND_PKTINFO);
 526
 527         lso_info = (struct ndis_tcp_lso_info *)((void *)ppi +
 528                         ppi->ppi_offset);
 529
 530         lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
 531         if (net_trans_info & (INFO_IPV4 << 16)) {
 532                 lso_info->lso_v2_transmit.ip_version =
 533                         NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
 534                 ip_hdr(skb)->tot_len = 0;
 535                 ip_hdr(skb)->check = 0;
 536                 tcp_hdr(skb)->check =
 537                 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
 538                                    ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 539         } else {
 540                 lso_info->lso_v2_transmit.ip_version =
 541                         NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
 542                 ipv6_hdr(skb)->payload_len = 0;
 543                 tcp_hdr(skb)->check =
 544                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 545                                 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 546         }
 547         lso_info->lso_v2_transmit.tcp_header_offset = hdr_offset;
 548         lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
 549
 550 do_send:
 551         /* Start filling in the page buffers with the rndis hdr */
 552         rndis_msg->msg_len += rndis_msg_size;
 553         packet->total_data_buflen = rndis_msg->msg_len;
 554         packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
 555                                         skb, &packet->page_buf[0]);
 556
 557         ret = netvsc_send(net_device_ctx->device_ctx, packet);
 558
 559 drop:
 560         if (ret == 0) {
 561                 net->stats.tx_bytes += skb_length;
 562                 net->stats.tx_packets++;
 563         } else {
 564                 kfree(packet);
 565                 if (ret != -EAGAIN) {
 566                         dev_kfree_skb_any(skb);
 567                         net->stats.tx_dropped++;
 568                 }
 569         }
 570
 571         return (ret == -EAGAIN) ? NETDEV_TX_BUSY : NETDEV_TX_OK;
 572 }
 573
 574 /*
 575  * netvsc_linkstatus_callback - Link up/down notification
 576  */
 577 void netvsc_linkstatus_callback(struct hv_device *device_obj,
 578                                 struct rndis_message *resp)
 579 {
 580         struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
 581         struct net_device *net;
 582         struct net_device_context *ndev_ctx;
 583         struct netvsc_device *net_device;
 584         struct rndis_device *rdev;
 585
 586         net_device = hv_get_drvdata(device_obj);
 587         rdev = net_device->extension;
 588
 589         switch (indicate->status) {
 590         case RNDIS_STATUS_MEDIA_CONNECT:
 591                 rdev->link_state = false;
 592                 break;
 593         case RNDIS_STATUS_MEDIA_DISCONNECT:
 594                 rdev->link_state = true;
 595                 break;
 596         case RNDIS_STATUS_NETWORK_CHANGE:
 597                 rdev->link_change = true;
 598                 break;
 599         default:
 600                 return;
 601         }
 602
 603         net = net_device->ndev;
 604
 605         if (!net || net->reg_state != NETREG_REGISTERED)
 606                 return;
 607
 608         ndev_ctx = netdev_priv(net);
 609         if (!rdev->link_state) {
 610                 schedule_delayed_work(&ndev_ctx->dwork, 0);
 611                 schedule_delayed_work(&ndev_ctx->dwork, msecs_to_jiffies(20));
 612         } else {
 613                 schedule_delayed_work(&ndev_ctx->dwork, 0);
 614         }
 615 }
 616
 617 /*
 618  * netvsc_recv_callback -  Callback when we receive a packet from the
 619  * "wire" on the specified device.
 620  */
 621 int netvsc_recv_callback(struct hv_device *device_obj,
 622                                 struct hv_netvsc_packet *packet,
 623                                 struct ndis_tcp_ip_checksum_info *csum_info)
 624 {
 625         struct net_device *net;
 626         struct sk_buff *skb;
 627
 628         net = ((struct netvsc_device *)hv_get_drvdata(device_obj))->ndev;
 629         if (!net || net->reg_state != NETREG_REGISTERED) {
 630                 packet->status = NVSP_STAT_FAIL;
 631                 return 0;
 632         }
 633
 634         /* Allocate a skb - TODO direct I/O to pages? */
 635         skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
 636         if (unlikely(!skb)) {
 637                 ++net->stats.rx_dropped;
 638                 packet->status = NVSP_STAT_FAIL;
 639                 return 0;
 640         }
 641
 642         /*
 643          * Copy to skb. This copy is needed here since the memory pointed by
 644          * hv_netvsc_packet cannot be deallocated
 645          */
 646         memcpy(skb_put(skb, packet->total_data_buflen), packet->data,
 647                 packet->total_data_buflen);
 648
 649         skb->protocol = eth_type_trans(skb, net);
 650         if (csum_info) {
 651                 /* We only look at the IP checksum here.
 652                  * Should we be dropping the packet if checksum
 653                  * failed? How do we deal with other checksums - TCP/UDP?
 654                  */
 655                 if (csum_info->receive.ip_checksum_succeeded)
 656                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 657                 else
 658                         skb->ip_summed = CHECKSUM_NONE;
 659         }
 660
 661         if (packet->vlan_tci & VLAN_TAG_PRESENT)
 662                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 663                                        packet->vlan_tci);
 664
 665         skb_record_rx_queue(skb, packet->channel->
 666                             offermsg.offer.sub_channel_index);
 667
 668         net->stats.rx_packets++;
 669         net->stats.rx_bytes += packet->total_data_buflen;
 670
 671         /*
 672          * Pass the skb back up. Network stack will deallocate the skb when it
 673          * is done.
 674          * TODO - use NAPI?
 675          */
 676         netif_rx(skb);
 677
 678         return 0;
 679 }
 680
 681 static void netvsc_get_drvinfo(struct net_device *net,
 682                                struct ethtool_drvinfo *info)
 683 {
 684         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
 685         strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
 686 }
 687
 688 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
 689 {
 690         struct net_device_context *ndevctx = netdev_priv(ndev);
 691         struct hv_device *hdev =  ndevctx->device_ctx;
 692         struct netvsc_device *nvdev = hv_get_drvdata(hdev);
 693         struct netvsc_device_info device_info;
 694         int limit = ETH_DATA_LEN;
 695
 696         if (nvdev == NULL || nvdev->destroy)
 697                 return -ENODEV;
 698
 699         if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
 700                 limit = NETVSC_MTU;
 701
 702         if (mtu < 68 || mtu > limit)
 703                 return -EINVAL;
 704
 705         nvdev->start_remove = true;
 706         cancel_work_sync(&ndevctx->work);
 707         netif_tx_disable(ndev);
 708         rndis_filter_device_remove(hdev);
 709
 710         ndev->mtu = mtu;
 711
 712         ndevctx->device_ctx = hdev;
 713         hv_set_drvdata(hdev, ndev);
 714         device_info.ring_size = ring_size;
 715         rndis_filter_device_add(hdev, &device_info);
 716         netif_tx_wake_all_queues(ndev);
 717
 718         return 0;
 719 }
 720
 721
 722 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
 723 {
 724         struct net_device_context *ndevctx = netdev_priv(ndev);
 725         struct hv_device *hdev =  ndevctx->device_ctx;
 726         struct sockaddr *addr = p;
 727         char save_adr[ETH_ALEN];
 728         unsigned char save_aatype;
 729         int err;
 730
 731         memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
 732         save_aatype = ndev->addr_assign_type;
 733
 734         err = eth_mac_addr(ndev, p);
 735         if (err != 0)
 736                 return err;
 737
 738         err = rndis_filter_set_device_mac(hdev, addr->sa_data);
 739         if (err != 0) {
 740                 /* roll back to saved MAC */
 741                 memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
 742                 ndev->addr_assign_type = save_aatype;
 743         }
 744
 745         return err;
 746 }
 747
 748 #ifdef CONFIG_NET_POLL_CONTROLLER
 749 static void netvsc_poll_controller(struct net_device *net)
 750 {
 751         /* As netvsc_start_xmit() works synchronous we don't have to
 752          * trigger anything here.
 753          */
 754 }
 755 #endif
 756
 757 static const struct ethtool_ops ethtool_ops = {
 758         .get_drvinfo    = netvsc_get_drvinfo,
 759         .get_link       = ethtool_op_get_link,
 760 };
 761
 762 static const struct net_device_ops device_ops = {
 763         .ndo_open =                     netvsc_open,
 764         .ndo_stop =                     netvsc_close,
 765         .ndo_start_xmit =               netvsc_start_xmit,
 766         .ndo_set_rx_mode =              netvsc_set_multicast_list,
 767         .ndo_change_mtu =               netvsc_change_mtu,
 768         .ndo_validate_addr =            eth_validate_addr,
 769         .ndo_set_mac_address =          netvsc_set_mac_addr,
 770         .ndo_select_queue =             netvsc_select_queue,
 771 #ifdef CONFIG_NET_POLL_CONTROLLER
 772         .ndo_poll_controller =          netvsc_poll_controller,
 773 #endif
 774 };
 775
 776 /*
 777  * Send GARP packet to network peers after migrations.
 778  * After Quick Migration, the network is not immediately operational in the
 779  * current context when receiving RNDIS_STATUS_MEDIA_CONNECT event. So, add
 780  * another netif_notify_peers() into a delayed work, otherwise GARP packet
 781  * will not be sent after quick migration, and cause network disconnection.
 782  * Also, we update the carrier status here.
 783  */
 784 static void netvsc_link_change(struct work_struct *w)
 785 {
 786         struct net_device_context *ndev_ctx;
 787         struct net_device *net;
 788         struct netvsc_device *net_device;
 789         struct rndis_device *rdev;
 790         bool notify, refresh = false;
 791         char *argv[] = { "/etc/init.d/network", "restart", NULL };
 792         char *envp[] = { "HOME=/", "PATH=/sbin:/usr/sbin:/bin:/usr/bin", NULL };
 793
 794         rtnl_lock();
 795
 796         ndev_ctx = container_of(w, struct net_device_context, dwork.work);
 797         net_device = hv_get_drvdata(ndev_ctx->device_ctx);
 798         rdev = net_device->extension;
 799         net = net_device->ndev;
 800
 801         if (rdev->link_state) {
 802                 netif_carrier_off(net);
 803                 notify = false;
 804         } else {
 805                 netif_carrier_on(net);
 806                 notify = true;
 807                 if (rdev->link_change) {
 808                         rdev->link_change = false;
 809                         refresh = true;
 810                 }
 811         }
 812
 813         rtnl_unlock();
 814
 815         if (refresh)
 816                 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
 817
 818         if (notify)
 819                 netdev_notify_peers(net);
 820 }
 821
 822
 823 static int netvsc_probe(struct hv_device *dev,
 824                         const struct hv_vmbus_device_id *dev_id)
 825 {
 826         struct net_device *net = NULL;
 827         struct net_device_context *net_device_ctx;
 828         struct netvsc_device_info device_info;
 829         struct netvsc_device *nvdev;
 830         int ret;
 831
 832         net = alloc_etherdev_mq(sizeof(struct net_device_context),
 833                                 num_online_cpus());
 834         if (!net)
 835                 return -ENOMEM;
 836
 837         netif_carrier_off(net);
 838
 839         net_device_ctx = netdev_priv(net);
 840         net_device_ctx->device_ctx = dev;
 841         hv_set_drvdata(dev, net);
 842         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
 843         INIT_WORK(&net_device_ctx->work, do_set_multicast);
 844
 845         net->netdev_ops = &device_ops;
 846
 847         net->hw_features = NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_IP_CSUM |
 848                                 NETIF_F_TSO;
 849         net->features = NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_SG | NETIF_F_RXCSUM |
 850                         NETIF_F_IP_CSUM | NETIF_F_TSO;
 851
 852         net->ethtool_ops = &ethtool_ops;
 853         SET_NETDEV_DEV(net, &dev->device);
 854
 855         /* Notify the netvsc driver of the new device */
 856         device_info.ring_size = ring_size;
 857         ret = rndis_filter_device_add(dev, &device_info);
 858         if (ret != 0) {
 859                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
 860                 free_netdev(net);
 861                 hv_set_drvdata(dev, NULL);
 862                 return ret;
 863         }
 864         memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
 865
 866         nvdev = hv_get_drvdata(dev);
 867         netif_set_real_num_tx_queues(net, nvdev->num_chn);
 868         netif_set_real_num_rx_queues(net, nvdev->num_chn);
 869
 870         ret = register_netdev(net);
 871         if (ret != 0) {
 872                 pr_err("Unable to register netdev.\n");
 873                 rndis_filter_device_remove(dev);
 874                 free_netdev(net);
 875         } else {
 876                 schedule_delayed_work(&net_device_ctx->dwork, 0);
 877         }
 878
 879         return ret;
 880 }
 881
 882 static int netvsc_remove(struct hv_device *dev)
 883 {
 884         struct net_device *net;
 885         struct net_device_context *ndev_ctx;
 886         struct netvsc_device *net_device;
 887
 888         net_device = hv_get_drvdata(dev);
 889         net = net_device->ndev;
 890
 891         if (net == NULL) {
 892                 dev_err(&dev->device, "No net device to remove\n");
 893                 return 0;
 894         }
 895
 896         net_device->start_remove = true;
 897
 898         ndev_ctx = netdev_priv(net);
 899         cancel_delayed_work_sync(&ndev_ctx->dwork);
 900         cancel_work_sync(&ndev_ctx->work);
 901
 902         /* Stop outbound asap */
 903         netif_tx_disable(net);
 904
 905         unregister_netdev(net);
 906
 907         /*
 908          * Call to the vsc driver to let it know that the device is being
 909          * removed
 910          */
 911         rndis_filter_device_remove(dev);
 912
 913         free_netdev(net);
 914         return 0;
 915 }
 916
 917 static const struct hv_vmbus_device_id id_table[] = {
 918         /* Network guid */
 919         { HV_NIC_GUID, },
 920         { },
 921 };
 922
 923 MODULE_DEVICE_TABLE(vmbus, id_table);
 924
 925 /* The one and only one */
 926 static struct  hv_driver netvsc_drv = {
 927         .name = KBUILD_MODNAME,
 928         .id_table = id_table,
 929         .probe = netvsc_probe,
 930         .remove = netvsc_remove,
 931 };
 932
 933 static void __exit netvsc_drv_exit(void)
 934 {
 935         vmbus_driver_unregister(&netvsc_drv);
 936 }
 937
 938 static int __init netvsc_drv_init(void)
 939 {
 940         if (ring_size < RING_SIZE_MIN) {
 941                 ring_size = RING_SIZE_MIN;
 942                 pr_info("Increased ring_size to %d (min allowed)\n",
 943                         ring_size);
 944         }
 945         return vmbus_driver_register(&netvsc_drv);
 946 }
 947
 948 MODULE_LICENSE("GPL");
 949 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
 950
 951 module_init(netvsc_drv_init);
 952 module_exit(netvsc_drv_exit);