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
  44 #define RING_SIZE_MIN 64
  45 static int ring_size = 128;
  46 module_param(ring_size, int, S_IRUGO);
  47 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
  48
  49 static int max_num_vrss_chns = 8;
  50
  51 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
  52                                 NETIF_MSG_LINK | NETIF_MSG_IFUP |
  53                                 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
  54                                 NETIF_MSG_TX_ERR;
  55
  56 static int debug = -1;
  57 module_param(debug, int, S_IRUGO);
  58 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  59
  60 static void do_set_multicast(struct work_struct *w)
  61 {
  62         struct net_device_context *ndevctx =
  63                 container_of(w, struct net_device_context, work);
  64         struct netvsc_device *nvdev;
  65         struct rndis_device *rdev;
  66
  67         nvdev = hv_get_drvdata(ndevctx->device_ctx);
  68         if (nvdev == NULL || nvdev->ndev == NULL)
  69                 return;
  70
  71         rdev = nvdev->extension;
  72         if (rdev == NULL)
  73                 return;
  74
  75         if (nvdev->ndev->flags & IFF_PROMISC)
  76                 rndis_filter_set_packet_filter(rdev,
  77                         NDIS_PACKET_TYPE_PROMISCUOUS);
  78         else
  79                 rndis_filter_set_packet_filter(rdev,
  80                         NDIS_PACKET_TYPE_BROADCAST |
  81                         NDIS_PACKET_TYPE_ALL_MULTICAST |
  82                         NDIS_PACKET_TYPE_DIRECTED);
  83 }
  84
  85 static void netvsc_set_multicast_list(struct net_device *net)
  86 {
  87         struct net_device_context *net_device_ctx = netdev_priv(net);
  88
  89         schedule_work(&net_device_ctx->work);
  90 }
  91
  92 static int netvsc_open(struct net_device *net)
  93 {
  94         struct net_device_context *net_device_ctx = netdev_priv(net);
  95         struct hv_device *device_obj = net_device_ctx->device_ctx;
  96         struct netvsc_device *nvdev;
  97         struct rndis_device *rdev;
  98         int ret = 0;
  99
 100         netif_carrier_off(net);
 101
 102         /* Open up the device */
 103         ret = rndis_filter_open(device_obj);
 104         if (ret != 0) {
 105                 netdev_err(net, "unable to open device (ret %d).\n", ret);
 106                 return ret;
 107         }
 108
 109         netif_tx_start_all_queues(net);
 110
 111         nvdev = hv_get_drvdata(device_obj);
 112         rdev = nvdev->extension;
 113         if (!rdev->link_state)
 114                 netif_carrier_on(net);
 115
 116         return ret;
 117 }
 118
 119 static int netvsc_close(struct net_device *net)
 120 {
 121         struct net_device_context *net_device_ctx = netdev_priv(net);
 122         struct hv_device *device_obj = net_device_ctx->device_ctx;
 123         int ret;
 124
 125         netif_tx_disable(net);
 126
 127         /* Make sure netvsc_set_multicast_list doesn't re-enable filter! */
 128         cancel_work_sync(&net_device_ctx->work);
 129         ret = rndis_filter_close(device_obj);
 130         if (ret != 0)
 131                 netdev_err(net, "unable to close device (ret %d).\n", ret);
 132
 133         return ret;
 134 }
 135
 136 static void *init_ppi_data(struct rndis_message *msg, u32 ppi_size,
 137                                 int pkt_type)
 138 {
 139         struct rndis_packet *rndis_pkt;
 140         struct rndis_per_packet_info *ppi;
 141
 142         rndis_pkt = &msg->msg.pkt;
 143         rndis_pkt->data_offset += ppi_size;
 144
 145         ppi = (struct rndis_per_packet_info *)((void *)rndis_pkt +
 146                 rndis_pkt->per_pkt_info_offset + rndis_pkt->per_pkt_info_len);
 147
 148         ppi->size = ppi_size;
 149         ppi->type = pkt_type;
 150         ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
 151
 152         rndis_pkt->per_pkt_info_len += ppi_size;
 153
 154         return ppi;
 155 }
 156
 157 union sub_key {
 158         u64 k;
 159         struct {
 160                 u8 pad[3];
 161                 u8 kb;
 162                 u32 ka;
 163         };
 164 };
 165
 166 /* Toeplitz hash function
 167  * data: network byte order
 168  * return: host byte order
 169  */
 170 static u32 comp_hash(u8 *key, int klen, void *data, int dlen)
 171 {
 172         union sub_key subk;
 173         int k_next = 4;
 174         u8 dt;
 175         int i, j;
 176         u32 ret = 0;
 177
 178         subk.k = 0;
 179         subk.ka = ntohl(*(u32 *)key);
 180
 181         for (i = 0; i < dlen; i++) {
 182                 subk.kb = key[k_next];
 183                 k_next = (k_next + 1) % klen;
 184                 dt = ((u8 *)data)[i];
 185                 for (j = 0; j < 8; j++) {
 186                         if (dt & 0x80)
 187                                 ret ^= subk.ka;
 188                         dt <<= 1;
 189                         subk.k <<= 1;
 190                 }
 191         }
 192
 193         return ret;
 194 }
 195
 196 static bool netvsc_set_hash(u32 *hash, struct sk_buff *skb)
 197 {
 198         struct flow_keys flow;
 199         int data_len;
 200
 201         if (!skb_flow_dissect_flow_keys(skb, &flow) ||
 202             !(flow.basic.n_proto == htons(ETH_P_IP) ||
 203               flow.basic.n_proto == htons(ETH_P_IPV6)))
 204                 return false;
 205
 206         if (flow.basic.ip_proto == IPPROTO_TCP)
 207                 data_len = 12;
 208         else
 209                 data_len = 8;
 210
 211         *hash = comp_hash(netvsc_hash_key, HASH_KEYLEN, &flow, data_len);
 212
 213         return true;
 214 }
 215
 216 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
 217                         void *accel_priv, select_queue_fallback_t fallback)
 218 {
 219         struct net_device_context *net_device_ctx = netdev_priv(ndev);
 220         struct hv_device *hdev =  net_device_ctx->device_ctx;
 221         struct netvsc_device *nvsc_dev = hv_get_drvdata(hdev);
 222         u32 hash;
 223         u16 q_idx = 0;
 224
 225         if (nvsc_dev == NULL || ndev->real_num_tx_queues <= 1)
 226                 return 0;
 227
 228         if (netvsc_set_hash(&hash, skb)) {
 229                 q_idx = nvsc_dev->send_table[hash % VRSS_SEND_TAB_SIZE] %
 230                         ndev->real_num_tx_queues;
 231                 skb_set_hash(skb, hash, PKT_HASH_TYPE_L3);
 232         }
 233
 234         return q_idx;
 235 }
 236
 237 void netvsc_xmit_completion(void *context)
 238 {
 239         struct hv_netvsc_packet *packet = (struct hv_netvsc_packet *)context;
 240         struct sk_buff *skb = (struct sk_buff *)
 241                 (unsigned long)packet->send_completion_tid;
 242
 243         if (skb)
 244                 dev_kfree_skb_any(skb);
 245 }
 246
 247 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
 248                         struct hv_page_buffer *pb)
 249 {
 250         int j = 0;
 251
 252         /* Deal with compund pages by ignoring unused part
 253          * of the page.
 254          */
 255         page += (offset >> PAGE_SHIFT);
 256         offset &= ~PAGE_MASK;
 257
 258         while (len > 0) {
 259                 unsigned long bytes;
 260
 261                 bytes = PAGE_SIZE - offset;
 262                 if (bytes > len)
 263                         bytes = len;
 264                 pb[j].pfn = page_to_pfn(page);
 265                 pb[j].offset = offset;
 266                 pb[j].len = bytes;
 267
 268                 offset += bytes;
 269                 len -= bytes;
 270
 271                 if (offset == PAGE_SIZE && len) {
 272                         page++;
 273                         offset = 0;
 274                         j++;
 275                 }
 276         }
 277
 278         return j + 1;
 279 }
 280
 281 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
 282                            struct hv_netvsc_packet *packet)
 283 {
 284         struct hv_page_buffer *pb = packet->page_buf;
 285         u32 slots_used = 0;
 286         char *data = skb->data;
 287         int frags = skb_shinfo(skb)->nr_frags;
 288         int i;
 289
 290         /* The packet is laid out thus:
 291          * 1. hdr: RNDIS header and PPI
 292          * 2. skb linear data
 293          * 3. skb fragment data
 294          */
 295         if (hdr != NULL)
 296                 slots_used += fill_pg_buf(virt_to_page(hdr),
 297                                         offset_in_page(hdr),
 298                                         len, &pb[slots_used]);
 299
 300         packet->rmsg_size = len;
 301         packet->rmsg_pgcnt = slots_used;
 302
 303         slots_used += fill_pg_buf(virt_to_page(data),
 304                                 offset_in_page(data),
 305                                 skb_headlen(skb), &pb[slots_used]);
 306
 307         for (i = 0; i < frags; i++) {
 308                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 309
 310                 slots_used += fill_pg_buf(skb_frag_page(frag),
 311                                         frag->page_offset,
 312                                         skb_frag_size(frag), &pb[slots_used]);
 313         }
 314         return slots_used;
 315 }
 316
 317 static int count_skb_frag_slots(struct sk_buff *skb)
 318 {
 319         int i, frags = skb_shinfo(skb)->nr_frags;
 320         int pages = 0;
 321
 322         for (i = 0; i < frags; i++) {
 323                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 324                 unsigned long size = skb_frag_size(frag);
 325                 unsigned long offset = frag->page_offset;
 326
 327                 /* Skip unused frames from start of page */
 328                 offset &= ~PAGE_MASK;
 329                 pages += PFN_UP(offset + size);
 330         }
 331         return pages;
 332 }
 333
 334 static int netvsc_get_slots(struct sk_buff *skb)
 335 {
 336         char *data = skb->data;
 337         unsigned int offset = offset_in_page(data);
 338         unsigned int len = skb_headlen(skb);
 339         int slots;
 340         int frag_slots;
 341
 342         slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
 343         frag_slots = count_skb_frag_slots(skb);
 344         return slots + frag_slots;
 345 }
 346
 347 static u32 get_net_transport_info(struct sk_buff *skb, u32 *trans_off)
 348 {
 349         u32 ret_val = TRANSPORT_INFO_NOT_IP;
 350
 351         if ((eth_hdr(skb)->h_proto != htons(ETH_P_IP)) &&
 352                 (eth_hdr(skb)->h_proto != htons(ETH_P_IPV6))) {
 353                 goto not_ip;
 354         }
 355
 356         *trans_off = skb_transport_offset(skb);
 357
 358         if ((eth_hdr(skb)->h_proto == htons(ETH_P_IP))) {
 359                 struct iphdr *iphdr = ip_hdr(skb);
 360
 361                 if (iphdr->protocol == IPPROTO_TCP)
 362                         ret_val = TRANSPORT_INFO_IPV4_TCP;
 363                 else if (iphdr->protocol == IPPROTO_UDP)
 364                         ret_val = TRANSPORT_INFO_IPV4_UDP;
 365         } else {
 366                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
 367                         ret_val = TRANSPORT_INFO_IPV6_TCP;
 368                 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
 369                         ret_val = TRANSPORT_INFO_IPV6_UDP;
 370         }
 371
 372 not_ip:
 373         return ret_val;
 374 }
 375
 376 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
 377 {
 378         struct net_device_context *net_device_ctx = netdev_priv(net);
 379         struct hv_netvsc_packet *packet = NULL;
 380         int ret;
 381         unsigned int num_data_pgs;
 382         struct rndis_message *rndis_msg;
 383         struct rndis_packet *rndis_pkt;
 384         u32 rndis_msg_size;
 385         bool isvlan;
 386         bool linear = false;
 387         struct rndis_per_packet_info *ppi;
 388         struct ndis_tcp_ip_checksum_info *csum_info;
 389         struct ndis_tcp_lso_info *lso_info;
 390         int  hdr_offset;
 391         u32 net_trans_info;
 392         u32 hash;
 393         u32 skb_length;
 394         u32 pkt_sz;
 395         struct hv_page_buffer page_buf[MAX_PAGE_BUFFER_COUNT];
 396         struct netvsc_stats *tx_stats = this_cpu_ptr(net_device_ctx->tx_stats);
 397
 398         /* We will atmost need two pages to describe the rndis
 399          * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
 400          * of pages in a single packet. If skb is scattered around
 401          * more pages we try linearizing it.
 402          */
 403
 404 check_size:
 405         skb_length = skb->len;
 406         num_data_pgs = netvsc_get_slots(skb) + 2;
 407         if (num_data_pgs > MAX_PAGE_BUFFER_COUNT && linear) {
 408                 net_alert_ratelimited("packet too big: %u pages (%u bytes)\n",
 409                                       num_data_pgs, skb->len);
 410                 ret = -EFAULT;
 411                 goto drop;
 412         } else if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
 413                 if (skb_linearize(skb)) {
 414                         net_alert_ratelimited("failed to linearize skb\n");
 415                         ret = -ENOMEM;
 416                         goto drop;
 417                 }
 418                 linear = true;
 419                 goto check_size;
 420         }
 421
 422         pkt_sz = sizeof(struct hv_netvsc_packet) + RNDIS_AND_PPI_SIZE;
 423
 424         ret = skb_cow_head(skb, pkt_sz);
 425         if (ret) {
 426                 netdev_err(net, "unable to alloc hv_netvsc_packet\n");
 427                 ret = -ENOMEM;
 428                 goto drop;
 429         }
 430         /* Use the headroom for building up the packet */
 431         packet = (struct hv_netvsc_packet *)skb->head;
 432
 433         packet->status = 0;
 434         packet->xmit_more = skb->xmit_more;
 435
 436         packet->vlan_tci = skb->vlan_tci;
 437         packet->page_buf = page_buf;
 438
 439         packet->q_idx = skb_get_queue_mapping(skb);
 440
 441         packet->is_data_pkt = true;
 442         packet->total_data_buflen = skb->len;
 443
 444         packet->rndis_msg = (struct rndis_message *)((unsigned long)packet +
 445                                 sizeof(struct hv_netvsc_packet));
 446
 447         memset(packet->rndis_msg, 0, RNDIS_AND_PPI_SIZE);
 448
 449         /* Set the completion routine */
 450         packet->send_completion = netvsc_xmit_completion;
 451         packet->send_completion_ctx = packet;
 452         packet->send_completion_tid = (unsigned long)skb;
 453
 454         isvlan = packet->vlan_tci & VLAN_TAG_PRESENT;
 455
 456         /* Add the rndis header */
 457         rndis_msg = packet->rndis_msg;
 458         rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
 459         rndis_msg->msg_len = packet->total_data_buflen;
 460         rndis_pkt = &rndis_msg->msg.pkt;
 461         rndis_pkt->data_offset = sizeof(struct rndis_packet);
 462         rndis_pkt->data_len = packet->total_data_buflen;
 463         rndis_pkt->per_pkt_info_offset = sizeof(struct rndis_packet);
 464
 465         rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
 466
 467         hash = skb_get_hash_raw(skb);
 468         if (hash != 0 && net->real_num_tx_queues > 1) {
 469                 rndis_msg_size += NDIS_HASH_PPI_SIZE;
 470                 ppi = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
 471                                     NBL_HASH_VALUE);
 472                 *(u32 *)((void *)ppi + ppi->ppi_offset) = hash;
 473         }
 474
 475         if (isvlan) {
 476                 struct ndis_pkt_8021q_info *vlan;
 477
 478                 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
 479                 ppi = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
 480                                         IEEE_8021Q_INFO);
 481                 vlan = (struct ndis_pkt_8021q_info *)((void *)ppi +
 482                                                 ppi->ppi_offset);
 483                 vlan->vlanid = packet->vlan_tci & VLAN_VID_MASK;
 484                 vlan->pri = (packet->vlan_tci & VLAN_PRIO_MASK) >>
 485                                 VLAN_PRIO_SHIFT;
 486         }
 487
 488         net_trans_info = get_net_transport_info(skb, &hdr_offset);
 489         if (net_trans_info == TRANSPORT_INFO_NOT_IP)
 490                 goto do_send;
 491
 492         /*
 493          * Setup the sendside checksum offload only if this is not a
 494          * GSO packet.
 495          */
 496         if (skb_is_gso(skb))
 497                 goto do_lso;
 498
 499         if ((skb->ip_summed == CHECKSUM_NONE) ||
 500             (skb->ip_summed == CHECKSUM_UNNECESSARY))
 501                 goto do_send;
 502
 503         rndis_msg_size += NDIS_CSUM_PPI_SIZE;
 504         ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
 505                             TCPIP_CHKSUM_PKTINFO);
 506
 507         csum_info = (struct ndis_tcp_ip_checksum_info *)((void *)ppi +
 508                         ppi->ppi_offset);
 509
 510         if (net_trans_info & (INFO_IPV4 << 16))
 511                 csum_info->transmit.is_ipv4 = 1;
 512         else
 513                 csum_info->transmit.is_ipv6 = 1;
 514
 515         if (net_trans_info & INFO_TCP) {
 516                 csum_info->transmit.tcp_checksum = 1;
 517                 csum_info->transmit.tcp_header_offset = hdr_offset;
 518         } else if (net_trans_info & INFO_UDP) {
 519                 /* UDP checksum offload is not supported on ws2008r2.
 520                  * Furthermore, on ws2012 and ws2012r2, there are some
 521                  * issues with udp checksum offload from Linux guests.
 522                  * (these are host issues).
 523                  * For now compute the checksum here.
 524                  */
 525                 struct udphdr *uh;
 526                 u16 udp_len;
 527
 528                 ret = skb_cow_head(skb, 0);
 529                 if (ret)
 530                         goto drop;
 531
 532                 uh = udp_hdr(skb);
 533                 udp_len = ntohs(uh->len);
 534                 uh->check = 0;
 535                 uh->check = csum_tcpudp_magic(ip_hdr(skb)->saddr,
 536                                               ip_hdr(skb)->daddr,
 537                                               udp_len, IPPROTO_UDP,
 538                                               csum_partial(uh, udp_len, 0));
 539                 if (uh->check == 0)
 540                         uh->check = CSUM_MANGLED_0;
 541
 542                 csum_info->transmit.udp_checksum = 0;
 543         }
 544         goto do_send;
 545
 546 do_lso:
 547         rndis_msg_size += NDIS_LSO_PPI_SIZE;
 548         ppi = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
 549                             TCP_LARGESEND_PKTINFO);
 550
 551         lso_info = (struct ndis_tcp_lso_info *)((void *)ppi +
 552                         ppi->ppi_offset);
 553
 554         lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
 555         if (net_trans_info & (INFO_IPV4 << 16)) {
 556                 lso_info->lso_v2_transmit.ip_version =
 557                         NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
 558                 ip_hdr(skb)->tot_len = 0;
 559                 ip_hdr(skb)->check = 0;
 560                 tcp_hdr(skb)->check =
 561                 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
 562                                    ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 563         } else {
 564                 lso_info->lso_v2_transmit.ip_version =
 565                         NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
 566                 ipv6_hdr(skb)->payload_len = 0;
 567                 tcp_hdr(skb)->check =
 568                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 569                                 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 570         }
 571         lso_info->lso_v2_transmit.tcp_header_offset = hdr_offset;
 572         lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
 573
 574 do_send:
 575         /* Start filling in the page buffers with the rndis hdr */
 576         rndis_msg->msg_len += rndis_msg_size;
 577         packet->total_data_buflen = rndis_msg->msg_len;
 578         packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
 579                                                skb, packet);
 580
 581         ret = netvsc_send(net_device_ctx->device_ctx, packet);
 582
 583 drop:
 584         if (ret == 0) {
 585                 u64_stats_update_begin(&tx_stats->syncp);
 586                 tx_stats->packets++;
 587                 tx_stats->bytes += skb_length;
 588                 u64_stats_update_end(&tx_stats->syncp);
 589         } else {
 590                 if (ret != -EAGAIN) {
 591                         dev_kfree_skb_any(skb);
 592                         net->stats.tx_dropped++;
 593                 }
 594         }
 595
 596         return (ret == -EAGAIN) ? NETDEV_TX_BUSY : NETDEV_TX_OK;
 597 }
 598
 599 /*
 600  * netvsc_linkstatus_callback - Link up/down notification
 601  */
 602 void netvsc_linkstatus_callback(struct hv_device *device_obj,
 603                                 struct rndis_message *resp)
 604 {
 605         struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
 606         struct net_device *net;
 607         struct net_device_context *ndev_ctx;
 608         struct netvsc_device *net_device;
 609         struct rndis_device *rdev;
 610
 611         net_device = hv_get_drvdata(device_obj);
 612         rdev = net_device->extension;
 613
 614         switch (indicate->status) {
 615         case RNDIS_STATUS_MEDIA_CONNECT:
 616                 rdev->link_state = false;
 617                 break;
 618         case RNDIS_STATUS_MEDIA_DISCONNECT:
 619                 rdev->link_state = true;
 620                 break;
 621         case RNDIS_STATUS_NETWORK_CHANGE:
 622                 rdev->link_change = true;
 623                 break;
 624         default:
 625                 return;
 626         }
 627
 628         net = net_device->ndev;
 629
 630         if (!net || net->reg_state != NETREG_REGISTERED)
 631                 return;
 632
 633         ndev_ctx = netdev_priv(net);
 634         if (!rdev->link_state) {
 635                 schedule_delayed_work(&ndev_ctx->dwork, 0);
 636                 schedule_delayed_work(&ndev_ctx->dwork, msecs_to_jiffies(20));
 637         } else {
 638                 schedule_delayed_work(&ndev_ctx->dwork, 0);
 639         }
 640 }
 641
 642 /*
 643  * netvsc_recv_callback -  Callback when we receive a packet from the
 644  * "wire" on the specified device.
 645  */
 646 int netvsc_recv_callback(struct hv_device *device_obj,
 647                                 struct hv_netvsc_packet *packet,
 648                                 struct ndis_tcp_ip_checksum_info *csum_info)
 649 {
 650         struct net_device *net;
 651         struct net_device_context *net_device_ctx;
 652         struct sk_buff *skb;
 653         struct netvsc_stats *rx_stats;
 654
 655         net = ((struct netvsc_device *)hv_get_drvdata(device_obj))->ndev;
 656         if (!net || net->reg_state != NETREG_REGISTERED) {
 657                 packet->status = NVSP_STAT_FAIL;
 658                 return 0;
 659         }
 660         net_device_ctx = netdev_priv(net);
 661         rx_stats = this_cpu_ptr(net_device_ctx->rx_stats);
 662
 663         /* Allocate a skb - TODO direct I/O to pages? */
 664         skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
 665         if (unlikely(!skb)) {
 666                 ++net->stats.rx_dropped;
 667                 packet->status = NVSP_STAT_FAIL;
 668                 return 0;
 669         }
 670
 671         /*
 672          * Copy to skb. This copy is needed here since the memory pointed by
 673          * hv_netvsc_packet cannot be deallocated
 674          */
 675         memcpy(skb_put(skb, packet->total_data_buflen), packet->data,
 676                 packet->total_data_buflen);
 677
 678         skb->protocol = eth_type_trans(skb, net);
 679         if (csum_info) {
 680                 /* We only look at the IP checksum here.
 681                  * Should we be dropping the packet if checksum
 682                  * failed? How do we deal with other checksums - TCP/UDP?
 683                  */
 684                 if (csum_info->receive.ip_checksum_succeeded)
 685                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 686                 else
 687                         skb->ip_summed = CHECKSUM_NONE;
 688         }
 689
 690         if (packet->vlan_tci & VLAN_TAG_PRESENT)
 691                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 692                                        packet->vlan_tci);
 693
 694         skb_record_rx_queue(skb, packet->channel->
 695                             offermsg.offer.sub_channel_index);
 696
 697         u64_stats_update_begin(&rx_stats->syncp);
 698         rx_stats->packets++;
 699         rx_stats->bytes += packet->total_data_buflen;
 700         u64_stats_update_end(&rx_stats->syncp);
 701
 702         /*
 703          * Pass the skb back up. Network stack will deallocate the skb when it
 704          * is done.
 705          * TODO - use NAPI?
 706          */
 707         netif_rx(skb);
 708
 709         return 0;
 710 }
 711
 712 static void netvsc_get_drvinfo(struct net_device *net,
 713                                struct ethtool_drvinfo *info)
 714 {
 715         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
 716         strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
 717 }
 718
 719 static void netvsc_get_channels(struct net_device *net,
 720                                 struct ethtool_channels *channel)
 721 {
 722         struct net_device_context *net_device_ctx = netdev_priv(net);
 723         struct hv_device *dev = net_device_ctx->device_ctx;
 724         struct netvsc_device *nvdev = hv_get_drvdata(dev);
 725
 726         if (nvdev) {
 727                 channel->max_combined   = nvdev->max_chn;
 728                 channel->combined_count = nvdev->num_chn;
 729         }
 730 }
 731
 732 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
 733 {
 734         struct net_device_context *ndevctx = netdev_priv(ndev);
 735         struct hv_device *hdev =  ndevctx->device_ctx;
 736         struct netvsc_device *nvdev = hv_get_drvdata(hdev);
 737         struct netvsc_device_info device_info;
 738         int limit = ETH_DATA_LEN;
 739
 740         if (nvdev == NULL || nvdev->destroy)
 741                 return -ENODEV;
 742
 743         if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
 744                 limit = NETVSC_MTU - ETH_HLEN;
 745
 746         /* Hyper-V hosts don't support MTU < ETH_DATA_LEN (1500) */
 747         if (mtu < ETH_DATA_LEN || mtu > limit)
 748                 return -EINVAL;
 749
 750         nvdev->start_remove = true;
 751         cancel_work_sync(&ndevctx->work);
 752         netif_tx_disable(ndev);
 753         rndis_filter_device_remove(hdev);
 754
 755         ndev->mtu = mtu;
 756
 757         ndevctx->device_ctx = hdev;
 758         hv_set_drvdata(hdev, ndev);
 759         device_info.ring_size = ring_size;
 760         device_info.max_num_vrss_chns = max_num_vrss_chns;
 761         rndis_filter_device_add(hdev, &device_info);
 762         netif_tx_wake_all_queues(ndev);
 763
 764         return 0;
 765 }
 766
 767 static struct rtnl_link_stats64 *netvsc_get_stats64(struct net_device *net,
 768                                                     struct rtnl_link_stats64 *t)
 769 {
 770         struct net_device_context *ndev_ctx = netdev_priv(net);
 771         int cpu;
 772
 773         for_each_possible_cpu(cpu) {
 774                 struct netvsc_stats *tx_stats = per_cpu_ptr(ndev_ctx->tx_stats,
 775                                                             cpu);
 776                 struct netvsc_stats *rx_stats = per_cpu_ptr(ndev_ctx->rx_stats,
 777                                                             cpu);
 778                 u64 tx_packets, tx_bytes, rx_packets, rx_bytes;
 779                 unsigned int start;
 780
 781                 do {
 782                         start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
 783                         tx_packets = tx_stats->packets;
 784                         tx_bytes = tx_stats->bytes;
 785                 } while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
 786
 787                 do {
 788                         start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
 789                         rx_packets = rx_stats->packets;
 790                         rx_bytes = rx_stats->bytes;
 791                 } while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
 792
 793                 t->tx_bytes     += tx_bytes;
 794                 t->tx_packets   += tx_packets;
 795                 t->rx_bytes     += rx_bytes;
 796                 t->rx_packets   += rx_packets;
 797         }
 798
 799         t->tx_dropped   = net->stats.tx_dropped;
 800         t->tx_errors    = net->stats.tx_dropped;
 801
 802         t->rx_dropped   = net->stats.rx_dropped;
 803         t->rx_errors    = net->stats.rx_errors;
 804
 805         return t;
 806 }
 807
 808 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
 809 {
 810         struct net_device_context *ndevctx = netdev_priv(ndev);
 811         struct hv_device *hdev =  ndevctx->device_ctx;
 812         struct sockaddr *addr = p;
 813         char save_adr[ETH_ALEN];
 814         unsigned char save_aatype;
 815         int err;
 816
 817         memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
 818         save_aatype = ndev->addr_assign_type;
 819
 820         err = eth_mac_addr(ndev, p);
 821         if (err != 0)
 822                 return err;
 823
 824         err = rndis_filter_set_device_mac(hdev, addr->sa_data);
 825         if (err != 0) {
 826                 /* roll back to saved MAC */
 827                 memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
 828                 ndev->addr_assign_type = save_aatype;
 829         }
 830
 831         return err;
 832 }
 833
 834 #ifdef CONFIG_NET_POLL_CONTROLLER
 835 static void netvsc_poll_controller(struct net_device *net)
 836 {
 837         /* As netvsc_start_xmit() works synchronous we don't have to
 838          * trigger anything here.
 839          */
 840 }
 841 #endif
 842
 843 static const struct ethtool_ops ethtool_ops = {
 844         .get_drvinfo    = netvsc_get_drvinfo,
 845         .get_link       = ethtool_op_get_link,
 846         .get_channels   = netvsc_get_channels,
 847 };
 848
 849 static const struct net_device_ops device_ops = {
 850         .ndo_open =                     netvsc_open,
 851         .ndo_stop =                     netvsc_close,
 852         .ndo_start_xmit =               netvsc_start_xmit,
 853         .ndo_set_rx_mode =              netvsc_set_multicast_list,
 854         .ndo_change_mtu =               netvsc_change_mtu,
 855         .ndo_validate_addr =            eth_validate_addr,
 856         .ndo_set_mac_address =          netvsc_set_mac_addr,
 857         .ndo_select_queue =             netvsc_select_queue,
 858         .ndo_get_stats64 =              netvsc_get_stats64,
 859 #ifdef CONFIG_NET_POLL_CONTROLLER
 860         .ndo_poll_controller =          netvsc_poll_controller,
 861 #endif
 862 };
 863
 864 /*
 865  * Send GARP packet to network peers after migrations.
 866  * After Quick Migration, the network is not immediately operational in the
 867  * current context when receiving RNDIS_STATUS_MEDIA_CONNECT event. So, add
 868  * another netif_notify_peers() into a delayed work, otherwise GARP packet
 869  * will not be sent after quick migration, and cause network disconnection.
 870  * Also, we update the carrier status here.
 871  */
 872 static void netvsc_link_change(struct work_struct *w)
 873 {
 874         struct net_device_context *ndev_ctx;
 875         struct net_device *net;
 876         struct netvsc_device *net_device;
 877         struct rndis_device *rdev;
 878         bool notify, refresh = false;
 879         char *argv[] = { "/etc/init.d/network", "restart", NULL };
 880         char *envp[] = { "HOME=/", "PATH=/sbin:/usr/sbin:/bin:/usr/bin", NULL };
 881
 882         rtnl_lock();
 883
 884         ndev_ctx = container_of(w, struct net_device_context, dwork.work);
 885         net_device = hv_get_drvdata(ndev_ctx->device_ctx);
 886         rdev = net_device->extension;
 887         net = net_device->ndev;
 888
 889         if (rdev->link_state) {
 890                 netif_carrier_off(net);
 891                 notify = false;
 892         } else {
 893                 netif_carrier_on(net);
 894                 notify = true;
 895                 if (rdev->link_change) {
 896                         rdev->link_change = false;
 897                         refresh = true;
 898                 }
 899         }
 900
 901         rtnl_unlock();
 902
 903         if (refresh)
 904                 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
 905
 906         if (notify)
 907                 netdev_notify_peers(net);
 908 }
 909
 910 static void netvsc_free_netdev(struct net_device *netdev)
 911 {
 912         struct net_device_context *net_device_ctx = netdev_priv(netdev);
 913
 914         free_percpu(net_device_ctx->tx_stats);
 915         free_percpu(net_device_ctx->rx_stats);
 916         free_netdev(netdev);
 917 }
 918
 919 static int netvsc_probe(struct hv_device *dev,
 920                         const struct hv_vmbus_device_id *dev_id)
 921 {
 922         struct net_device *net = NULL;
 923         struct net_device_context *net_device_ctx;
 924         struct netvsc_device_info device_info;
 925         struct netvsc_device *nvdev;
 926         int ret;
 927         u32 max_needed_headroom;
 928
 929         net = alloc_etherdev_mq(sizeof(struct net_device_context),
 930                                 num_online_cpus());
 931         if (!net)
 932                 return -ENOMEM;
 933
 934         max_needed_headroom = sizeof(struct hv_netvsc_packet) +
 935                               RNDIS_AND_PPI_SIZE;
 936
 937         netif_carrier_off(net);
 938
 939         net_device_ctx = netdev_priv(net);
 940         net_device_ctx->device_ctx = dev;
 941         net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
 942         if (netif_msg_probe(net_device_ctx))
 943                 netdev_dbg(net, "netvsc msg_enable: %d\n",
 944                            net_device_ctx->msg_enable);
 945
 946         net_device_ctx->tx_stats = netdev_alloc_pcpu_stats(struct netvsc_stats);
 947         if (!net_device_ctx->tx_stats) {
 948                 free_netdev(net);
 949                 return -ENOMEM;
 950         }
 951         net_device_ctx->rx_stats = netdev_alloc_pcpu_stats(struct netvsc_stats);
 952         if (!net_device_ctx->rx_stats) {
 953                 free_percpu(net_device_ctx->tx_stats);
 954                 free_netdev(net);
 955                 return -ENOMEM;
 956         }
 957
 958         hv_set_drvdata(dev, net);
 959         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
 960         INIT_WORK(&net_device_ctx->work, do_set_multicast);
 961
 962         net->netdev_ops = &device_ops;
 963
 964         net->hw_features = NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_IP_CSUM |
 965                                 NETIF_F_TSO;
 966         net->features = NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_SG | NETIF_F_RXCSUM |
 967                         NETIF_F_IP_CSUM | NETIF_F_TSO;
 968
 969         net->ethtool_ops = &ethtool_ops;
 970         SET_NETDEV_DEV(net, &dev->device);
 971
 972         /*
 973          * Request additional head room in the skb.
 974          * We will use this space to build the rndis
 975          * heaser and other state we need to maintain.
 976          */
 977         net->needed_headroom = max_needed_headroom;
 978
 979         /* Notify the netvsc driver of the new device */
 980         device_info.ring_size = ring_size;
 981         device_info.max_num_vrss_chns = max_num_vrss_chns;
 982         ret = rndis_filter_device_add(dev, &device_info);
 983         if (ret != 0) {
 984                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
 985                 netvsc_free_netdev(net);
 986                 hv_set_drvdata(dev, NULL);
 987                 return ret;
 988         }
 989         memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
 990
 991         nvdev = hv_get_drvdata(dev);
 992         netif_set_real_num_tx_queues(net, nvdev->num_chn);
 993         netif_set_real_num_rx_queues(net, nvdev->num_chn);
 994
 995         ret = register_netdev(net);
 996         if (ret != 0) {
 997                 pr_err("Unable to register netdev.\n");
 998                 rndis_filter_device_remove(dev);
 999                 netvsc_free_netdev(net);
1000         } else {
1001                 schedule_delayed_work(&net_device_ctx->dwork, 0);
1002         }
1003
1004         return ret;
1005 }
1006
1007 static int netvsc_remove(struct hv_device *dev)
1008 {
1009         struct net_device *net;
1010         struct net_device_context *ndev_ctx;
1011         struct netvsc_device *net_device;
1012
1013         net_device = hv_get_drvdata(dev);
1014         net = net_device->ndev;
1015
1016         if (net == NULL) {
1017                 dev_err(&dev->device, "No net device to remove\n");
1018                 return 0;
1019         }
1020
1021         net_device->start_remove = true;
1022
1023         ndev_ctx = netdev_priv(net);
1024         cancel_delayed_work_sync(&ndev_ctx->dwork);
1025         cancel_work_sync(&ndev_ctx->work);
1026
1027         /* Stop outbound asap */
1028         netif_tx_disable(net);
1029
1030         unregister_netdev(net);
1031
1032         /*
1033          * Call to the vsc driver to let it know that the device is being
1034          * removed
1035          */
1036         rndis_filter_device_remove(dev);
1037
1038         netvsc_free_netdev(net);
1039         return 0;
1040 }
1041
1042 static const struct hv_vmbus_device_id id_table[] = {
1043         /* Network guid */
1044         { HV_NIC_GUID, },
1045         { },
1046 };
1047
1048 MODULE_DEVICE_TABLE(vmbus, id_table);
1049
1050 /* The one and only one */
1051 static struct  hv_driver netvsc_drv = {
1052         .name = KBUILD_MODNAME,
1053         .id_table = id_table,
1054         .probe = netvsc_probe,
1055         .remove = netvsc_remove,
1056 };
1057
1058 static void __exit netvsc_drv_exit(void)
1059 {
1060         vmbus_driver_unregister(&netvsc_drv);
1061 }
1062
1063 static int __init netvsc_drv_init(void)
1064 {
1065         if (ring_size < RING_SIZE_MIN) {
1066                 ring_size = RING_SIZE_MIN;
1067                 pr_info("Increased ring_size to %d (min allowed)\n",
1068                         ring_size);
1069         }
1070         return vmbus_driver_register(&netvsc_drv);
1071 }
1072
1073 MODULE_LICENSE("GPL");
1074 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
1075
1076 module_init(netvsc_drv_init);
1077 module_exit(netvsc_drv_exit);