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 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
 238         if (!packet->part_of_skb)
 239                 kfree(packet);
 240
 241         if (skb)
 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 = NULL;
 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         u32 head_room = skb_headroom(skb);
 388         u32 pkt_sz;
 389         struct hv_page_buffer page_buf[MAX_PAGE_BUFFER_COUNT];
 390
 391
 392         /* We will atmost need two pages to describe the rndis
 393          * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
 394          * of pages in a single packet.
 395          */
 396         num_data_pgs = netvsc_get_slots(skb) + 2;
 397         if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
 398                 netdev_err(net, "Packet too big: %u\n", skb->len);
 399                 ret = -EFAULT;
 400                 goto drop;
 401         }
 402
 403         pkt_sz = sizeof(struct hv_netvsc_packet) + RNDIS_AND_PPI_SIZE;
 404
 405         if (head_room < pkt_sz) {
 406                 packet = kmalloc(pkt_sz, GFP_ATOMIC);
 407                 if (!packet) {
 408                         /* out of memory, drop packet */
 409                         netdev_err(net, "unable to alloc hv_netvsc_packet\n");
 410                         ret = -ENOMEM;
 411                         goto drop;
 412                 }
 413                 packet->part_of_skb = false;
 414         } else {
 415                 /* Use the headroom for building up the packet */
 416                 packet = (struct hv_netvsc_packet *)skb->head;
 417                 packet->part_of_skb = true;
 418         }
 419
 420         packet->status = 0;
 421         packet->xmit_more = skb->xmit_more;
 422
 423         packet->vlan_tci = skb->vlan_tci;
 424         packet->page_buf = page_buf;
 425
 426         packet->q_idx = skb_get_queue_mapping(skb);
 427
 428         packet->is_data_pkt = true;
 429         packet->total_data_buflen = skb->len;
 430
 431         packet->rndis_msg = (struct rndis_message *)((unsigned long)packet +
 432                                 sizeof(struct hv_netvsc_packet));
 433
 434         memset(packet->rndis_msg, 0, RNDIS_AND_PPI_SIZE);
 435
 436         /* Set the completion routine */
 437         packet->send_completion = netvsc_xmit_completion;
 438         packet->send_completion_ctx = packet;
 439         packet->send_completion_tid = (unsigned long)skb;
 440
 441         isvlan = packet->vlan_tci & VLAN_TAG_PRESENT;
 442
 443         /* Add the rndis header */
 444         rndis_msg = packet->rndis_msg;
 445         rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
 446         rndis_msg->msg_len = packet->total_data_buflen;
 447         rndis_pkt = &rndis_msg->msg.pkt;
 448         rndis_pkt->data_offset = sizeof(struct rndis_packet);
 449         rndis_pkt->data_len = packet->total_data_buflen;
 450         rndis_pkt->per_pkt_info_offset = sizeof(struct rndis_packet);
 451
 452         rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
 453
 454         hash = skb_get_hash_raw(skb);
 455         if (hash != 0 && net->real_num_tx_queues > 1) {
 456                 rndis_msg_size += NDIS_HASH_PPI_SIZE;
 457                 ppi = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
 458                                     NBL_HASH_VALUE);
 459                 *(u32 *)((void *)ppi + ppi->ppi_offset) = hash;
 460         }
 461
 462         if (isvlan) {
 463                 struct ndis_pkt_8021q_info *vlan;
 464
 465                 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
 466                 ppi = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
 467                                         IEEE_8021Q_INFO);
 468                 vlan = (struct ndis_pkt_8021q_info *)((void *)ppi +
 469                                                 ppi->ppi_offset);
 470                 vlan->vlanid = packet->vlan_tci & VLAN_VID_MASK;
 471                 vlan->pri = (packet->vlan_tci & VLAN_PRIO_MASK) >>
 472                                 VLAN_PRIO_SHIFT;
 473         }
 474
 475         net_trans_info = get_net_transport_info(skb, &hdr_offset);
 476         if (net_trans_info == TRANSPORT_INFO_NOT_IP)
 477                 goto do_send;
 478
 479         /*
 480          * Setup the sendside checksum offload only if this is not a
 481          * GSO packet.
 482          */
 483         if (skb_is_gso(skb))
 484                 goto do_lso;
 485
 486         if ((skb->ip_summed == CHECKSUM_NONE) ||
 487             (skb->ip_summed == CHECKSUM_UNNECESSARY))
 488                 goto do_send;
 489
 490         rndis_msg_size += NDIS_CSUM_PPI_SIZE;
 491         ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
 492                             TCPIP_CHKSUM_PKTINFO);
 493
 494         csum_info = (struct ndis_tcp_ip_checksum_info *)((void *)ppi +
 495                         ppi->ppi_offset);
 496
 497         if (net_trans_info & (INFO_IPV4 << 16))
 498                 csum_info->transmit.is_ipv4 = 1;
 499         else
 500                 csum_info->transmit.is_ipv6 = 1;
 501
 502         if (net_trans_info & INFO_TCP) {
 503                 csum_info->transmit.tcp_checksum = 1;
 504                 csum_info->transmit.tcp_header_offset = hdr_offset;
 505         } else if (net_trans_info & INFO_UDP) {
 506                 /* UDP checksum offload is not supported on ws2008r2.
 507                  * Furthermore, on ws2012 and ws2012r2, there are some
 508                  * issues with udp checksum offload from Linux guests.
 509                  * (these are host issues).
 510                  * For now compute the checksum here.
 511                  */
 512                 struct udphdr *uh;
 513                 u16 udp_len;
 514
 515                 ret = skb_cow_head(skb, 0);
 516                 if (ret)
 517                         goto drop;
 518
 519                 uh = udp_hdr(skb);
 520                 udp_len = ntohs(uh->len);
 521                 uh->check = 0;
 522                 uh->check = csum_tcpudp_magic(ip_hdr(skb)->saddr,
 523                                               ip_hdr(skb)->daddr,
 524                                               udp_len, IPPROTO_UDP,
 525                                               csum_partial(uh, udp_len, 0));
 526                 if (uh->check == 0)
 527                         uh->check = CSUM_MANGLED_0;
 528
 529                 csum_info->transmit.udp_checksum = 0;
 530         }
 531         goto do_send;
 532
 533 do_lso:
 534         rndis_msg_size += NDIS_LSO_PPI_SIZE;
 535         ppi = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
 536                             TCP_LARGESEND_PKTINFO);
 537
 538         lso_info = (struct ndis_tcp_lso_info *)((void *)ppi +
 539                         ppi->ppi_offset);
 540
 541         lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
 542         if (net_trans_info & (INFO_IPV4 << 16)) {
 543                 lso_info->lso_v2_transmit.ip_version =
 544                         NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
 545                 ip_hdr(skb)->tot_len = 0;
 546                 ip_hdr(skb)->check = 0;
 547                 tcp_hdr(skb)->check =
 548                 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
 549                                    ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 550         } else {
 551                 lso_info->lso_v2_transmit.ip_version =
 552                         NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
 553                 ipv6_hdr(skb)->payload_len = 0;
 554                 tcp_hdr(skb)->check =
 555                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 556                                 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 557         }
 558         lso_info->lso_v2_transmit.tcp_header_offset = hdr_offset;
 559         lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
 560
 561 do_send:
 562         /* Start filling in the page buffers with the rndis hdr */
 563         rndis_msg->msg_len += rndis_msg_size;
 564         packet->total_data_buflen = rndis_msg->msg_len;
 565         packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
 566                                         skb, &page_buf[0]);
 567
 568         ret = netvsc_send(net_device_ctx->device_ctx, packet);
 569
 570 drop:
 571         if (ret == 0) {
 572                 net->stats.tx_bytes += skb_length;
 573                 net->stats.tx_packets++;
 574         } else {
 575                 if (packet && !packet->part_of_skb)
 576                         kfree(packet);
 577                 if (ret != -EAGAIN) {
 578                         dev_kfree_skb_any(skb);
 579                         net->stats.tx_dropped++;
 580                 }
 581         }
 582
 583         return (ret == -EAGAIN) ? NETDEV_TX_BUSY : NETDEV_TX_OK;
 584 }
 585
 586 /*
 587  * netvsc_linkstatus_callback - Link up/down notification
 588  */
 589 void netvsc_linkstatus_callback(struct hv_device *device_obj,
 590                                 struct rndis_message *resp)
 591 {
 592         struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
 593         struct net_device *net;
 594         struct net_device_context *ndev_ctx;
 595         struct netvsc_device *net_device;
 596         struct rndis_device *rdev;
 597
 598         net_device = hv_get_drvdata(device_obj);
 599         rdev = net_device->extension;
 600
 601         switch (indicate->status) {
 602         case RNDIS_STATUS_MEDIA_CONNECT:
 603                 rdev->link_state = false;
 604                 break;
 605         case RNDIS_STATUS_MEDIA_DISCONNECT:
 606                 rdev->link_state = true;
 607                 break;
 608         case RNDIS_STATUS_NETWORK_CHANGE:
 609                 rdev->link_change = true;
 610                 break;
 611         default:
 612                 return;
 613         }
 614
 615         net = net_device->ndev;
 616
 617         if (!net || net->reg_state != NETREG_REGISTERED)
 618                 return;
 619
 620         ndev_ctx = netdev_priv(net);
 621         if (!rdev->link_state) {
 622                 schedule_delayed_work(&ndev_ctx->dwork, 0);
 623                 schedule_delayed_work(&ndev_ctx->dwork, msecs_to_jiffies(20));
 624         } else {
 625                 schedule_delayed_work(&ndev_ctx->dwork, 0);
 626         }
 627 }
 628
 629 /*
 630  * netvsc_recv_callback -  Callback when we receive a packet from the
 631  * "wire" on the specified device.
 632  */
 633 int netvsc_recv_callback(struct hv_device *device_obj,
 634                                 struct hv_netvsc_packet *packet,
 635                                 struct ndis_tcp_ip_checksum_info *csum_info)
 636 {
 637         struct net_device *net;
 638         struct sk_buff *skb;
 639
 640         net = ((struct netvsc_device *)hv_get_drvdata(device_obj))->ndev;
 641         if (!net || net->reg_state != NETREG_REGISTERED) {
 642                 packet->status = NVSP_STAT_FAIL;
 643                 return 0;
 644         }
 645
 646         /* Allocate a skb - TODO direct I/O to pages? */
 647         skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
 648         if (unlikely(!skb)) {
 649                 ++net->stats.rx_dropped;
 650                 packet->status = NVSP_STAT_FAIL;
 651                 return 0;
 652         }
 653
 654         /*
 655          * Copy to skb. This copy is needed here since the memory pointed by
 656          * hv_netvsc_packet cannot be deallocated
 657          */
 658         memcpy(skb_put(skb, packet->total_data_buflen), packet->data,
 659                 packet->total_data_buflen);
 660
 661         skb->protocol = eth_type_trans(skb, net);
 662         if (csum_info) {
 663                 /* We only look at the IP checksum here.
 664                  * Should we be dropping the packet if checksum
 665                  * failed? How do we deal with other checksums - TCP/UDP?
 666                  */
 667                 if (csum_info->receive.ip_checksum_succeeded)
 668                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 669                 else
 670                         skb->ip_summed = CHECKSUM_NONE;
 671         }
 672
 673         if (packet->vlan_tci & VLAN_TAG_PRESENT)
 674                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 675                                        packet->vlan_tci);
 676
 677         skb_record_rx_queue(skb, packet->channel->
 678                             offermsg.offer.sub_channel_index);
 679
 680         net->stats.rx_packets++;
 681         net->stats.rx_bytes += packet->total_data_buflen;
 682
 683         /*
 684          * Pass the skb back up. Network stack will deallocate the skb when it
 685          * is done.
 686          * TODO - use NAPI?
 687          */
 688         netif_rx(skb);
 689
 690         return 0;
 691 }
 692
 693 static void netvsc_get_drvinfo(struct net_device *net,
 694                                struct ethtool_drvinfo *info)
 695 {
 696         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
 697         strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
 698 }
 699
 700 static void netvsc_get_channels(struct net_device *net,
 701                                 struct ethtool_channels *channel)
 702 {
 703         struct net_device_context *net_device_ctx = netdev_priv(net);
 704         struct hv_device *dev = net_device_ctx->device_ctx;
 705         struct netvsc_device *nvdev = hv_get_drvdata(dev);
 706
 707         if (nvdev) {
 708                 channel->max_combined   = nvdev->max_chn;
 709                 channel->combined_count = nvdev->num_chn;
 710         }
 711 }
 712
 713 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
 714 {
 715         struct net_device_context *ndevctx = netdev_priv(ndev);
 716         struct hv_device *hdev =  ndevctx->device_ctx;
 717         struct netvsc_device *nvdev = hv_get_drvdata(hdev);
 718         struct netvsc_device_info device_info;
 719         int limit = ETH_DATA_LEN;
 720
 721         if (nvdev == NULL || nvdev->destroy)
 722                 return -ENODEV;
 723
 724         if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
 725                 limit = NETVSC_MTU - ETH_HLEN;
 726
 727         /* Hyper-V hosts don't support MTU < ETH_DATA_LEN (1500) */
 728         if (mtu < ETH_DATA_LEN || mtu > limit)
 729                 return -EINVAL;
 730
 731         nvdev->start_remove = true;
 732         cancel_work_sync(&ndevctx->work);
 733         netif_tx_disable(ndev);
 734         rndis_filter_device_remove(hdev);
 735
 736         ndev->mtu = mtu;
 737
 738         ndevctx->device_ctx = hdev;
 739         hv_set_drvdata(hdev, ndev);
 740         device_info.ring_size = ring_size;
 741         rndis_filter_device_add(hdev, &device_info);
 742         netif_tx_wake_all_queues(ndev);
 743
 744         return 0;
 745 }
 746
 747
 748 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
 749 {
 750         struct net_device_context *ndevctx = netdev_priv(ndev);
 751         struct hv_device *hdev =  ndevctx->device_ctx;
 752         struct sockaddr *addr = p;
 753         char save_adr[ETH_ALEN];
 754         unsigned char save_aatype;
 755         int err;
 756
 757         memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
 758         save_aatype = ndev->addr_assign_type;
 759
 760         err = eth_mac_addr(ndev, p);
 761         if (err != 0)
 762                 return err;
 763
 764         err = rndis_filter_set_device_mac(hdev, addr->sa_data);
 765         if (err != 0) {
 766                 /* roll back to saved MAC */
 767                 memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
 768                 ndev->addr_assign_type = save_aatype;
 769         }
 770
 771         return err;
 772 }
 773
 774 #ifdef CONFIG_NET_POLL_CONTROLLER
 775 static void netvsc_poll_controller(struct net_device *net)
 776 {
 777         /* As netvsc_start_xmit() works synchronous we don't have to
 778          * trigger anything here.
 779          */
 780 }
 781 #endif
 782
 783 static const struct ethtool_ops ethtool_ops = {
 784         .get_drvinfo    = netvsc_get_drvinfo,
 785         .get_link       = ethtool_op_get_link,
 786         .get_channels   = netvsc_get_channels,
 787 };
 788
 789 static const struct net_device_ops device_ops = {
 790         .ndo_open =                     netvsc_open,
 791         .ndo_stop =                     netvsc_close,
 792         .ndo_start_xmit =               netvsc_start_xmit,
 793         .ndo_set_rx_mode =              netvsc_set_multicast_list,
 794         .ndo_change_mtu =               netvsc_change_mtu,
 795         .ndo_validate_addr =            eth_validate_addr,
 796         .ndo_set_mac_address =          netvsc_set_mac_addr,
 797         .ndo_select_queue =             netvsc_select_queue,
 798 #ifdef CONFIG_NET_POLL_CONTROLLER
 799         .ndo_poll_controller =          netvsc_poll_controller,
 800 #endif
 801 };
 802
 803 /*
 804  * Send GARP packet to network peers after migrations.
 805  * After Quick Migration, the network is not immediately operational in the
 806  * current context when receiving RNDIS_STATUS_MEDIA_CONNECT event. So, add
 807  * another netif_notify_peers() into a delayed work, otherwise GARP packet
 808  * will not be sent after quick migration, and cause network disconnection.
 809  * Also, we update the carrier status here.
 810  */
 811 static void netvsc_link_change(struct work_struct *w)
 812 {
 813         struct net_device_context *ndev_ctx;
 814         struct net_device *net;
 815         struct netvsc_device *net_device;
 816         struct rndis_device *rdev;
 817         bool notify, refresh = false;
 818         char *argv[] = { "/etc/init.d/network", "restart", NULL };
 819         char *envp[] = { "HOME=/", "PATH=/sbin:/usr/sbin:/bin:/usr/bin", NULL };
 820
 821         rtnl_lock();
 822
 823         ndev_ctx = container_of(w, struct net_device_context, dwork.work);
 824         net_device = hv_get_drvdata(ndev_ctx->device_ctx);
 825         rdev = net_device->extension;
 826         net = net_device->ndev;
 827
 828         if (rdev->link_state) {
 829                 netif_carrier_off(net);
 830                 notify = false;
 831         } else {
 832                 netif_carrier_on(net);
 833                 notify = true;
 834                 if (rdev->link_change) {
 835                         rdev->link_change = false;
 836                         refresh = true;
 837                 }
 838         }
 839
 840         rtnl_unlock();
 841
 842         if (refresh)
 843                 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
 844
 845         if (notify)
 846                 netdev_notify_peers(net);
 847 }
 848
 849
 850 static int netvsc_probe(struct hv_device *dev,
 851                         const struct hv_vmbus_device_id *dev_id)
 852 {
 853         struct net_device *net = NULL;
 854         struct net_device_context *net_device_ctx;
 855         struct netvsc_device_info device_info;
 856         struct netvsc_device *nvdev;
 857         int ret;
 858         u32 max_needed_headroom;
 859
 860         net = alloc_etherdev_mq(sizeof(struct net_device_context),
 861                                 num_online_cpus());
 862         if (!net)
 863                 return -ENOMEM;
 864
 865         max_needed_headroom = sizeof(struct hv_netvsc_packet) +
 866                               RNDIS_AND_PPI_SIZE;
 867
 868         netif_carrier_off(net);
 869
 870         net_device_ctx = netdev_priv(net);
 871         net_device_ctx->device_ctx = dev;
 872         hv_set_drvdata(dev, net);
 873         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
 874         INIT_WORK(&net_device_ctx->work, do_set_multicast);
 875
 876         net->netdev_ops = &device_ops;
 877
 878         net->hw_features = NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_IP_CSUM |
 879                                 NETIF_F_TSO;
 880         net->features = NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_SG | NETIF_F_RXCSUM |
 881                         NETIF_F_IP_CSUM | NETIF_F_TSO;
 882
 883         net->ethtool_ops = &ethtool_ops;
 884         SET_NETDEV_DEV(net, &dev->device);
 885
 886         /*
 887          * Request additional head room in the skb.
 888          * We will use this space to build the rndis
 889          * heaser and other state we need to maintain.
 890          */
 891         net->needed_headroom = max_needed_headroom;
 892
 893         /* Notify the netvsc driver of the new device */
 894         device_info.ring_size = ring_size;
 895         ret = rndis_filter_device_add(dev, &device_info);
 896         if (ret != 0) {
 897                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
 898                 free_netdev(net);
 899                 hv_set_drvdata(dev, NULL);
 900                 return ret;
 901         }
 902         memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
 903
 904         nvdev = hv_get_drvdata(dev);
 905         netif_set_real_num_tx_queues(net, nvdev->num_chn);
 906         netif_set_real_num_rx_queues(net, nvdev->num_chn);
 907
 908         ret = register_netdev(net);
 909         if (ret != 0) {
 910                 pr_err("Unable to register netdev.\n");
 911                 rndis_filter_device_remove(dev);
 912                 free_netdev(net);
 913         } else {
 914                 schedule_delayed_work(&net_device_ctx->dwork, 0);
 915         }
 916
 917         return ret;
 918 }
 919
 920 static int netvsc_remove(struct hv_device *dev)
 921 {
 922         struct net_device *net;
 923         struct net_device_context *ndev_ctx;
 924         struct netvsc_device *net_device;
 925
 926         net_device = hv_get_drvdata(dev);
 927         net = net_device->ndev;
 928
 929         if (net == NULL) {
 930                 dev_err(&dev->device, "No net device to remove\n");
 931                 return 0;
 932         }
 933
 934         net_device->start_remove = true;
 935
 936         ndev_ctx = netdev_priv(net);
 937         cancel_delayed_work_sync(&ndev_ctx->dwork);
 938         cancel_work_sync(&ndev_ctx->work);
 939
 940         /* Stop outbound asap */
 941         netif_tx_disable(net);
 942
 943         unregister_netdev(net);
 944
 945         /*
 946          * Call to the vsc driver to let it know that the device is being
 947          * removed
 948          */
 949         rndis_filter_device_remove(dev);
 950
 951         free_netdev(net);
 952         return 0;
 953 }
 954
 955 static const struct hv_vmbus_device_id id_table[] = {
 956         /* Network guid */
 957         { HV_NIC_GUID, },
 958         { },
 959 };
 960
 961 MODULE_DEVICE_TABLE(vmbus, id_table);
 962
 963 /* The one and only one */
 964 static struct  hv_driver netvsc_drv = {
 965         .name = KBUILD_MODNAME,
 966         .id_table = id_table,
 967         .probe = netvsc_probe,
 968         .remove = netvsc_remove,
 969 };
 970
 971 static void __exit netvsc_drv_exit(void)
 972 {
 973         vmbus_driver_unregister(&netvsc_drv);
 974 }
 975
 976 static int __init netvsc_drv_init(void)
 977 {
 978         if (ring_size < RING_SIZE_MIN) {
 979                 ring_size = RING_SIZE_MIN;
 980                 pr_info("Increased ring_size to %d (min allowed)\n",
 981                         ring_size);
 982         }
 983         return vmbus_driver_register(&netvsc_drv);
 984 }
 985
 986 MODULE_LICENSE("GPL");
 987 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
 988
 989 module_init(netvsc_drv_init);
 990 module_exit(netvsc_drv_exit);