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                 skb_set_hash(skb, hash, PKT_HASH_TYPE_L3);
 231         }
 232
 233         return q_idx;
 234 }
 235
 236 static void netvsc_xmit_completion(void *context)
 237 {
 238         struct hv_netvsc_packet *packet = (struct hv_netvsc_packet *)context;
 239         struct sk_buff *skb = (struct sk_buff *)
 240                 (unsigned long)packet->send_completion_tid;
 241         u32 index = packet->send_buf_index;
 242
 243         kfree(packet);
 244
 245         if (skb && (index == NETVSC_INVALID_INDEX))
 246                 dev_kfree_skb_any(skb);
 247 }
 248
 249 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
 250                         struct hv_page_buffer *pb)
 251 {
 252         int j = 0;
 253
 254         /* Deal with compund pages by ignoring unused part
 255          * of the page.
 256          */
 257         page += (offset >> PAGE_SHIFT);
 258         offset &= ~PAGE_MASK;
 259
 260         while (len > 0) {
 261                 unsigned long bytes;
 262
 263                 bytes = PAGE_SIZE - offset;
 264                 if (bytes > len)
 265                         bytes = len;
 266                 pb[j].pfn = page_to_pfn(page);
 267                 pb[j].offset = offset;
 268                 pb[j].len = bytes;
 269
 270                 offset += bytes;
 271                 len -= bytes;
 272
 273                 if (offset == PAGE_SIZE && len) {
 274                         page++;
 275                         offset = 0;
 276                         j++;
 277                 }
 278         }
 279
 280         return j + 1;
 281 }
 282
 283 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
 284                            struct hv_page_buffer *pb)
 285 {
 286         u32 slots_used = 0;
 287         char *data = skb->data;
 288         int frags = skb_shinfo(skb)->nr_frags;
 289         int i;
 290
 291         /* The packet is laid out thus:
 292          * 1. hdr
 293          * 2. skb linear data
 294          * 3. skb fragment data
 295          */
 296         if (hdr != NULL)
 297                 slots_used += fill_pg_buf(virt_to_page(hdr),
 298                                         offset_in_page(hdr),
 299                                         len, &pb[slots_used]);
 300
 301         slots_used += fill_pg_buf(virt_to_page(data),
 302                                 offset_in_page(data),
 303                                 skb_headlen(skb), &pb[slots_used]);
 304
 305         for (i = 0; i < frags; i++) {
 306                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 307
 308                 slots_used += fill_pg_buf(skb_frag_page(frag),
 309                                         frag->page_offset,
 310                                         skb_frag_size(frag), &pb[slots_used]);
 311         }
 312         return slots_used;
 313 }
 314
 315 static int count_skb_frag_slots(struct sk_buff *skb)
 316 {
 317         int i, frags = skb_shinfo(skb)->nr_frags;
 318         int pages = 0;
 319
 320         for (i = 0; i < frags; i++) {
 321                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 322                 unsigned long size = skb_frag_size(frag);
 323                 unsigned long offset = frag->page_offset;
 324
 325                 /* Skip unused frames from start of page */
 326                 offset &= ~PAGE_MASK;
 327                 pages += PFN_UP(offset + size);
 328         }
 329         return pages;
 330 }
 331
 332 static int netvsc_get_slots(struct sk_buff *skb)
 333 {
 334         char *data = skb->data;
 335         unsigned int offset = offset_in_page(data);
 336         unsigned int len = skb_headlen(skb);
 337         int slots;
 338         int frag_slots;
 339
 340         slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
 341         frag_slots = count_skb_frag_slots(skb);
 342         return slots + frag_slots;
 343 }
 344
 345 static u32 get_net_transport_info(struct sk_buff *skb, u32 *trans_off)
 346 {
 347         u32 ret_val = TRANSPORT_INFO_NOT_IP;
 348
 349         if ((eth_hdr(skb)->h_proto != htons(ETH_P_IP)) &&
 350                 (eth_hdr(skb)->h_proto != htons(ETH_P_IPV6))) {
 351                 goto not_ip;
 352         }
 353
 354         *trans_off = skb_transport_offset(skb);
 355
 356         if ((eth_hdr(skb)->h_proto == htons(ETH_P_IP))) {
 357                 struct iphdr *iphdr = ip_hdr(skb);
 358
 359                 if (iphdr->protocol == IPPROTO_TCP)
 360                         ret_val = TRANSPORT_INFO_IPV4_TCP;
 361                 else if (iphdr->protocol == IPPROTO_UDP)
 362                         ret_val = TRANSPORT_INFO_IPV4_UDP;
 363         } else {
 364                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
 365                         ret_val = TRANSPORT_INFO_IPV6_TCP;
 366                 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
 367                         ret_val = TRANSPORT_INFO_IPV6_UDP;
 368         }
 369
 370 not_ip:
 371         return ret_val;
 372 }
 373
 374 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
 375 {
 376         struct net_device_context *net_device_ctx = netdev_priv(net);
 377         struct hv_netvsc_packet *packet;
 378         int ret;
 379         unsigned int num_data_pgs;
 380         struct rndis_message *rndis_msg;
 381         struct rndis_packet *rndis_pkt;
 382         u32 rndis_msg_size;
 383         bool isvlan;
 384         struct rndis_per_packet_info *ppi;
 385         struct ndis_tcp_ip_checksum_info *csum_info;
 386         struct ndis_tcp_lso_info *lso_info;
 387         int  hdr_offset;
 388         u32 net_trans_info;
 389         u32 hash;
 390         u32 skb_length = skb->len;
 391
 392
 393         /* We will atmost need two pages to describe the rndis
 394          * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
 395          * of pages in a single packet.
 396          */
 397         num_data_pgs = netvsc_get_slots(skb) + 2;
 398         if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
 399                 netdev_err(net, "Packet too big: %u\n", skb->len);
 400                 dev_kfree_skb(skb);
 401                 net->stats.tx_dropped++;
 402                 return NETDEV_TX_OK;
 403         }
 404
 405         /* Allocate a netvsc packet based on # of frags. */
 406         packet = kzalloc(sizeof(struct hv_netvsc_packet) +
 407                          (num_data_pgs * sizeof(struct hv_page_buffer)) +
 408                          sizeof(struct rndis_message) +
 409                          NDIS_VLAN_PPI_SIZE + NDIS_CSUM_PPI_SIZE +
 410                          NDIS_LSO_PPI_SIZE + NDIS_HASH_PPI_SIZE, GFP_ATOMIC);
 411         if (!packet) {
 412                 /* out of memory, drop packet */
 413                 netdev_err(net, "unable to allocate hv_netvsc_packet\n");
 414
 415                 dev_kfree_skb(skb);
 416                 net->stats.tx_dropped++;
 417                 return NETDEV_TX_OK;
 418         }
 419
 420         packet->vlan_tci = skb->vlan_tci;
 421
 422         packet->q_idx = skb_get_queue_mapping(skb);
 423
 424         packet->is_data_pkt = true;
 425         packet->total_data_buflen = skb->len;
 426
 427         packet->rndis_msg = (struct rndis_message *)((unsigned long)packet +
 428                                 sizeof(struct hv_netvsc_packet) +
 429                                 (num_data_pgs * sizeof(struct hv_page_buffer)));
 430
 431         /* Set the completion routine */
 432         packet->send_completion = netvsc_xmit_completion;
 433         packet->send_completion_ctx = packet;
 434         packet->send_completion_tid = (unsigned long)skb;
 435
 436         isvlan = packet->vlan_tci & VLAN_TAG_PRESENT;
 437
 438         /* Add the rndis header */
 439         rndis_msg = packet->rndis_msg;
 440         rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
 441         rndis_msg->msg_len = packet->total_data_buflen;
 442         rndis_pkt = &rndis_msg->msg.pkt;
 443         rndis_pkt->data_offset = sizeof(struct rndis_packet);
 444         rndis_pkt->data_len = packet->total_data_buflen;
 445         rndis_pkt->per_pkt_info_offset = sizeof(struct rndis_packet);
 446
 447         rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
 448
 449         hash = skb_get_hash_raw(skb);
 450         if (hash != 0 && net->real_num_tx_queues > 1) {
 451                 rndis_msg_size += NDIS_HASH_PPI_SIZE;
 452                 ppi = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
 453                                     NBL_HASH_VALUE);
 454                 *(u32 *)((void *)ppi + ppi->ppi_offset) = hash;
 455         }
 456
 457         if (isvlan) {
 458                 struct ndis_pkt_8021q_info *vlan;
 459
 460                 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
 461                 ppi = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
 462                                         IEEE_8021Q_INFO);
 463                 vlan = (struct ndis_pkt_8021q_info *)((void *)ppi +
 464                                                 ppi->ppi_offset);
 465                 vlan->vlanid = packet->vlan_tci & VLAN_VID_MASK;
 466                 vlan->pri = (packet->vlan_tci & VLAN_PRIO_MASK) >>
 467                                 VLAN_PRIO_SHIFT;
 468         }
 469
 470         net_trans_info = get_net_transport_info(skb, &hdr_offset);
 471         if (net_trans_info == TRANSPORT_INFO_NOT_IP)
 472                 goto do_send;
 473
 474         /*
 475          * Setup the sendside checksum offload only if this is not a
 476          * GSO packet.
 477          */
 478         if (skb_is_gso(skb))
 479                 goto do_lso;
 480
 481         if ((skb->ip_summed == CHECKSUM_NONE) ||
 482             (skb->ip_summed == CHECKSUM_UNNECESSARY))
 483                 goto do_send;
 484
 485         rndis_msg_size += NDIS_CSUM_PPI_SIZE;
 486         ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
 487                             TCPIP_CHKSUM_PKTINFO);
 488
 489         csum_info = (struct ndis_tcp_ip_checksum_info *)((void *)ppi +
 490                         ppi->ppi_offset);
 491
 492         if (net_trans_info & (INFO_IPV4 << 16))
 493                 csum_info->transmit.is_ipv4 = 1;
 494         else
 495                 csum_info->transmit.is_ipv6 = 1;
 496
 497         if (net_trans_info & INFO_TCP) {
 498                 csum_info->transmit.tcp_checksum = 1;
 499                 csum_info->transmit.tcp_header_offset = hdr_offset;
 500         } else if (net_trans_info & INFO_UDP) {
 501                 /* UDP checksum offload is not supported on ws2008r2.
 502                  * Furthermore, on ws2012 and ws2012r2, there are some
 503                  * issues with udp checksum offload from Linux guests.
 504                  * (these are host issues).
 505                  * For now compute the checksum here.
 506                  */
 507                 struct udphdr *uh;
 508                 u16 udp_len;
 509
 510                 ret = skb_cow_head(skb, 0);
 511                 if (ret)
 512                         goto drop;
 513
 514                 uh = udp_hdr(skb);
 515                 udp_len = ntohs(uh->len);
 516                 uh->check = 0;
 517                 uh->check = csum_tcpudp_magic(ip_hdr(skb)->saddr,
 518                                               ip_hdr(skb)->daddr,
 519                                               udp_len, IPPROTO_UDP,
 520                                               csum_partial(uh, udp_len, 0));
 521                 if (uh->check == 0)
 522                         uh->check = CSUM_MANGLED_0;
 523
 524                 csum_info->transmit.udp_checksum = 0;
 525         }
 526         goto do_send;
 527
 528 do_lso:
 529         rndis_msg_size += NDIS_LSO_PPI_SIZE;
 530         ppi = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
 531                             TCP_LARGESEND_PKTINFO);
 532
 533         lso_info = (struct ndis_tcp_lso_info *)((void *)ppi +
 534                         ppi->ppi_offset);
 535
 536         lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
 537         if (net_trans_info & (INFO_IPV4 << 16)) {
 538                 lso_info->lso_v2_transmit.ip_version =
 539                         NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
 540                 ip_hdr(skb)->tot_len = 0;
 541                 ip_hdr(skb)->check = 0;
 542                 tcp_hdr(skb)->check =
 543                 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
 544                                    ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 545         } else {
 546                 lso_info->lso_v2_transmit.ip_version =
 547                         NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
 548                 ipv6_hdr(skb)->payload_len = 0;
 549                 tcp_hdr(skb)->check =
 550                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 551                                 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 552         }
 553         lso_info->lso_v2_transmit.tcp_header_offset = hdr_offset;
 554         lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
 555
 556 do_send:
 557         /* Start filling in the page buffers with the rndis hdr */
 558         rndis_msg->msg_len += rndis_msg_size;
 559         packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
 560                                         skb, &packet->page_buf[0]);
 561
 562         ret = netvsc_send(net_device_ctx->device_ctx, packet);
 563
 564 drop:
 565         if (ret == 0) {
 566                 net->stats.tx_bytes += skb_length;
 567                 net->stats.tx_packets++;
 568         } else {
 569                 kfree(packet);
 570                 if (ret != -EAGAIN) {
 571                         dev_kfree_skb_any(skb);
 572                         net->stats.tx_dropped++;
 573                 }
 574         }
 575
 576         return (ret == -EAGAIN) ? NETDEV_TX_BUSY : NETDEV_TX_OK;
 577 }
 578
 579 /*
 580  * netvsc_linkstatus_callback - Link up/down notification
 581  */
 582 void netvsc_linkstatus_callback(struct hv_device *device_obj,
 583                                 struct rndis_message *resp)
 584 {
 585         struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
 586         struct net_device *net;
 587         struct net_device_context *ndev_ctx;
 588         struct netvsc_device *net_device;
 589         struct rndis_device *rdev;
 590
 591         net_device = hv_get_drvdata(device_obj);
 592         rdev = net_device->extension;
 593
 594         switch (indicate->status) {
 595         case RNDIS_STATUS_MEDIA_CONNECT:
 596                 rdev->link_state = false;
 597                 break;
 598         case RNDIS_STATUS_MEDIA_DISCONNECT:
 599                 rdev->link_state = true;
 600                 break;
 601         case RNDIS_STATUS_NETWORK_CHANGE:
 602                 rdev->link_change = true;
 603                 break;
 604         default:
 605                 return;
 606         }
 607
 608         net = net_device->ndev;
 609
 610         if (!net || net->reg_state != NETREG_REGISTERED)
 611                 return;
 612
 613         ndev_ctx = netdev_priv(net);
 614         if (!rdev->link_state) {
 615                 schedule_delayed_work(&ndev_ctx->dwork, 0);
 616                 schedule_delayed_work(&ndev_ctx->dwork, msecs_to_jiffies(20));
 617         } else {
 618                 schedule_delayed_work(&ndev_ctx->dwork, 0);
 619         }
 620 }
 621
 622 /*
 623  * netvsc_recv_callback -  Callback when we receive a packet from the
 624  * "wire" on the specified device.
 625  */
 626 int netvsc_recv_callback(struct hv_device *device_obj,
 627                                 struct hv_netvsc_packet *packet,
 628                                 struct ndis_tcp_ip_checksum_info *csum_info)
 629 {
 630         struct net_device *net;
 631         struct sk_buff *skb;
 632
 633         net = ((struct netvsc_device *)hv_get_drvdata(device_obj))->ndev;
 634         if (!net || net->reg_state != NETREG_REGISTERED) {
 635                 packet->status = NVSP_STAT_FAIL;
 636                 return 0;
 637         }
 638
 639         /* Allocate a skb - TODO direct I/O to pages? */
 640         skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
 641         if (unlikely(!skb)) {
 642                 ++net->stats.rx_dropped;
 643                 packet->status = NVSP_STAT_FAIL;
 644                 return 0;
 645         }
 646
 647         /*
 648          * Copy to skb. This copy is needed here since the memory pointed by
 649          * hv_netvsc_packet cannot be deallocated
 650          */
 651         memcpy(skb_put(skb, packet->total_data_buflen), packet->data,
 652                 packet->total_data_buflen);
 653
 654         skb->protocol = eth_type_trans(skb, net);
 655         if (csum_info) {
 656                 /* We only look at the IP checksum here.
 657                  * Should we be dropping the packet if checksum
 658                  * failed? How do we deal with other checksums - TCP/UDP?
 659                  */
 660                 if (csum_info->receive.ip_checksum_succeeded)
 661                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 662                 else
 663                         skb->ip_summed = CHECKSUM_NONE;
 664         }
 665
 666         if (packet->vlan_tci & VLAN_TAG_PRESENT)
 667                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 668                                        packet->vlan_tci);
 669
 670         skb_record_rx_queue(skb, packet->channel->
 671                             offermsg.offer.sub_channel_index);
 672
 673         net->stats.rx_packets++;
 674         net->stats.rx_bytes += packet->total_data_buflen;
 675
 676         /*
 677          * Pass the skb back up. Network stack will deallocate the skb when it
 678          * is done.
 679          * TODO - use NAPI?
 680          */
 681         netif_rx(skb);
 682
 683         return 0;
 684 }
 685
 686 static void netvsc_get_drvinfo(struct net_device *net,
 687                                struct ethtool_drvinfo *info)
 688 {
 689         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
 690         strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
 691 }
 692
 693 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
 694 {
 695         struct net_device_context *ndevctx = netdev_priv(ndev);
 696         struct hv_device *hdev =  ndevctx->device_ctx;
 697         struct netvsc_device *nvdev = hv_get_drvdata(hdev);
 698         struct netvsc_device_info device_info;
 699         int limit = ETH_DATA_LEN;
 700
 701         if (nvdev == NULL || nvdev->destroy)
 702                 return -ENODEV;
 703
 704         if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
 705                 limit = NETVSC_MTU;
 706
 707         if (mtu < 68 || mtu > limit)
 708                 return -EINVAL;
 709
 710         nvdev->start_remove = true;
 711         cancel_work_sync(&ndevctx->work);
 712         netif_tx_disable(ndev);
 713         rndis_filter_device_remove(hdev);
 714
 715         ndev->mtu = mtu;
 716
 717         ndevctx->device_ctx = hdev;
 718         hv_set_drvdata(hdev, ndev);
 719         device_info.ring_size = ring_size;
 720         rndis_filter_device_add(hdev, &device_info);
 721         netif_tx_wake_all_queues(ndev);
 722
 723         return 0;
 724 }
 725
 726
 727 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
 728 {
 729         struct net_device_context *ndevctx = netdev_priv(ndev);
 730         struct hv_device *hdev =  ndevctx->device_ctx;
 731         struct sockaddr *addr = p;
 732         char save_adr[ETH_ALEN];
 733         unsigned char save_aatype;
 734         int err;
 735
 736         memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
 737         save_aatype = ndev->addr_assign_type;
 738
 739         err = eth_mac_addr(ndev, p);
 740         if (err != 0)
 741                 return err;
 742
 743         err = rndis_filter_set_device_mac(hdev, addr->sa_data);
 744         if (err != 0) {
 745                 /* roll back to saved MAC */
 746                 memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
 747                 ndev->addr_assign_type = save_aatype;
 748         }
 749
 750         return err;
 751 }
 752
 753 #ifdef CONFIG_NET_POLL_CONTROLLER
 754 static void netvsc_poll_controller(struct net_device *net)
 755 {
 756         /* As netvsc_start_xmit() works synchronous we don't have to
 757          * trigger anything here.
 758          */
 759 }
 760 #endif
 761
 762 static const struct ethtool_ops ethtool_ops = {
 763         .get_drvinfo    = netvsc_get_drvinfo,
 764         .get_link       = ethtool_op_get_link,
 765 };
 766
 767 static const struct net_device_ops device_ops = {
 768         .ndo_open =                     netvsc_open,
 769         .ndo_stop =                     netvsc_close,
 770         .ndo_start_xmit =               netvsc_start_xmit,
 771         .ndo_set_rx_mode =              netvsc_set_multicast_list,
 772         .ndo_change_mtu =               netvsc_change_mtu,
 773         .ndo_validate_addr =            eth_validate_addr,
 774         .ndo_set_mac_address =          netvsc_set_mac_addr,
 775         .ndo_select_queue =             netvsc_select_queue,
 776 #ifdef CONFIG_NET_POLL_CONTROLLER
 777         .ndo_poll_controller =          netvsc_poll_controller,
 778 #endif
 779 };
 780
 781 /*
 782  * Send GARP packet to network peers after migrations.
 783  * After Quick Migration, the network is not immediately operational in the
 784  * current context when receiving RNDIS_STATUS_MEDIA_CONNECT event. So, add
 785  * another netif_notify_peers() into a delayed work, otherwise GARP packet
 786  * will not be sent after quick migration, and cause network disconnection.
 787  * Also, we update the carrier status here.
 788  */
 789 static void netvsc_link_change(struct work_struct *w)
 790 {
 791         struct net_device_context *ndev_ctx;
 792         struct net_device *net;
 793         struct netvsc_device *net_device;
 794         struct rndis_device *rdev;
 795         bool notify, refresh = false;
 796         char *argv[] = { "/etc/init.d/network", "restart", NULL };
 797         char *envp[] = { "HOME=/", "PATH=/sbin:/usr/sbin:/bin:/usr/bin", NULL };
 798
 799         rtnl_lock();
 800
 801         ndev_ctx = container_of(w, struct net_device_context, dwork.work);
 802         net_device = hv_get_drvdata(ndev_ctx->device_ctx);
 803         rdev = net_device->extension;
 804         net = net_device->ndev;
 805
 806         if (rdev->link_state) {
 807                 netif_carrier_off(net);
 808                 notify = false;
 809         } else {
 810                 netif_carrier_on(net);
 811                 notify = true;
 812                 if (rdev->link_change) {
 813                         rdev->link_change = false;
 814                         refresh = true;
 815                 }
 816         }
 817
 818         rtnl_unlock();
 819
 820         if (refresh)
 821                 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
 822
 823         if (notify)
 824                 netdev_notify_peers(net);
 825 }
 826
 827
 828 static int netvsc_probe(struct hv_device *dev,
 829                         const struct hv_vmbus_device_id *dev_id)
 830 {
 831         struct net_device *net = NULL;
 832         struct net_device_context *net_device_ctx;
 833         struct netvsc_device_info device_info;
 834         struct netvsc_device *nvdev;
 835         int ret;
 836
 837         net = alloc_etherdev_mq(sizeof(struct net_device_context),
 838                                 num_online_cpus());
 839         if (!net)
 840                 return -ENOMEM;
 841
 842         netif_carrier_off(net);
 843
 844         net_device_ctx = netdev_priv(net);
 845         net_device_ctx->device_ctx = dev;
 846         hv_set_drvdata(dev, net);
 847         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
 848         INIT_WORK(&net_device_ctx->work, do_set_multicast);
 849
 850         net->netdev_ops = &device_ops;
 851
 852         net->hw_features = NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_IP_CSUM |
 853                                 NETIF_F_TSO;
 854         net->features = NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_SG | NETIF_F_RXCSUM |
 855                         NETIF_F_IP_CSUM | NETIF_F_TSO;
 856
 857         net->ethtool_ops = &ethtool_ops;
 858         SET_NETDEV_DEV(net, &dev->device);
 859
 860         /* Notify the netvsc driver of the new device */
 861         device_info.ring_size = ring_size;
 862         ret = rndis_filter_device_add(dev, &device_info);
 863         if (ret != 0) {
 864                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
 865                 free_netdev(net);
 866                 hv_set_drvdata(dev, NULL);
 867                 return ret;
 868         }
 869         memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
 870
 871         nvdev = hv_get_drvdata(dev);
 872         netif_set_real_num_tx_queues(net, nvdev->num_chn);
 873         netif_set_real_num_rx_queues(net, nvdev->num_chn);
 874
 875         ret = register_netdev(net);
 876         if (ret != 0) {
 877                 pr_err("Unable to register netdev.\n");
 878                 rndis_filter_device_remove(dev);
 879                 free_netdev(net);
 880         } else {
 881                 schedule_delayed_work(&net_device_ctx->dwork, 0);
 882         }
 883
 884         return ret;
 885 }
 886
 887 static int netvsc_remove(struct hv_device *dev)
 888 {
 889         struct net_device *net;
 890         struct net_device_context *ndev_ctx;
 891         struct netvsc_device *net_device;
 892
 893         net_device = hv_get_drvdata(dev);
 894         net = net_device->ndev;
 895
 896         if (net == NULL) {
 897                 dev_err(&dev->device, "No net device to remove\n");
 898                 return 0;
 899         }
 900
 901         net_device->start_remove = true;
 902
 903         ndev_ctx = netdev_priv(net);
 904         cancel_delayed_work_sync(&ndev_ctx->dwork);
 905         cancel_work_sync(&ndev_ctx->work);
 906
 907         /* Stop outbound asap */
 908         netif_tx_disable(net);
 909
 910         unregister_netdev(net);
 911
 912         /*
 913          * Call to the vsc driver to let it know that the device is being
 914          * removed
 915          */
 916         rndis_filter_device_remove(dev);
 917
 918         free_netdev(net);
 919         return 0;
 920 }
 921
 922 static const struct hv_vmbus_device_id id_table[] = {
 923         /* Network guid */
 924         { HV_NIC_GUID, },
 925         { },
 926 };
 927
 928 MODULE_DEVICE_TABLE(vmbus, id_table);
 929
 930 /* The one and only one */
 931 static struct  hv_driver netvsc_drv = {
 932         .name = KBUILD_MODNAME,
 933         .id_table = id_table,
 934         .probe = netvsc_probe,
 935         .remove = netvsc_remove,
 936 };
 937
 938 static void __exit netvsc_drv_exit(void)
 939 {
 940         vmbus_driver_unregister(&netvsc_drv);
 941 }
 942
 943 static int __init netvsc_drv_init(void)
 944 {
 945         if (ring_size < RING_SIZE_MIN) {
 946                 ring_size = RING_SIZE_MIN;
 947                 pr_info("Increased ring_size to %d (min allowed)\n",
 948                         ring_size);
 949         }
 950         return vmbus_driver_register(&netvsc_drv);
 951 }
 952
 953 MODULE_LICENSE("GPL");
 954 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
 955
 956 module_init(netvsc_drv_init);
 957 module_exit(netvsc_drv_exit);