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 /* Restrict GSO size to account for NVGRE */
  44 #define NETVSC_GSO_MAX_SIZE     62768
  45
  46 #define RING_SIZE_MIN 64
  47 static int ring_size = 128;
  48 module_param(ring_size, int, S_IRUGO);
  49 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
  50
  51 static int max_num_vrss_chns = 8;
  52
  53 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
  54                                 NETIF_MSG_LINK | NETIF_MSG_IFUP |
  55                                 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
  56                                 NETIF_MSG_TX_ERR;
  57
  58 static int debug = -1;
  59 module_param(debug, int, S_IRUGO);
  60 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  61
  62 static void do_set_multicast(struct work_struct *w)
  63 {
  64         struct net_device_context *ndevctx =
  65                 container_of(w, struct net_device_context, work);
  66         struct netvsc_device *nvdev;
  67         struct rndis_device *rdev;
  68
  69         nvdev = hv_get_drvdata(ndevctx->device_ctx);
  70         if (nvdev == NULL || nvdev->ndev == NULL)
  71                 return;
  72
  73         rdev = nvdev->extension;
  74         if (rdev == NULL)
  75                 return;
  76
  77         if (nvdev->ndev->flags & IFF_PROMISC)
  78                 rndis_filter_set_packet_filter(rdev,
  79                         NDIS_PACKET_TYPE_PROMISCUOUS);
  80         else
  81                 rndis_filter_set_packet_filter(rdev,
  82                         NDIS_PACKET_TYPE_BROADCAST |
  83                         NDIS_PACKET_TYPE_ALL_MULTICAST |
  84                         NDIS_PACKET_TYPE_DIRECTED);
  85 }
  86
  87 static void netvsc_set_multicast_list(struct net_device *net)
  88 {
  89         struct net_device_context *net_device_ctx = netdev_priv(net);
  90
  91         schedule_work(&net_device_ctx->work);
  92 }
  93
  94 static int netvsc_open(struct net_device *net)
  95 {
  96         struct net_device_context *net_device_ctx = netdev_priv(net);
  97         struct hv_device *device_obj = net_device_ctx->device_ctx;
  98         struct netvsc_device *nvdev;
  99         struct rndis_device *rdev;
 100         int ret = 0;
 101
 102         netif_carrier_off(net);
 103
 104         /* Open up the device */
 105         ret = rndis_filter_open(device_obj);
 106         if (ret != 0) {
 107                 netdev_err(net, "unable to open device (ret %d).\n", ret);
 108                 return ret;
 109         }
 110
 111         netif_tx_wake_all_queues(net);
 112
 113         nvdev = hv_get_drvdata(device_obj);
 114         rdev = nvdev->extension;
 115         if (!rdev->link_state)
 116                 netif_carrier_on(net);
 117
 118         return ret;
 119 }
 120
 121 static int netvsc_close(struct net_device *net)
 122 {
 123         struct net_device_context *net_device_ctx = netdev_priv(net);
 124         struct hv_device *device_obj = net_device_ctx->device_ctx;
 125         struct netvsc_device *nvdev = hv_get_drvdata(device_obj);
 126         int ret;
 127         u32 aread, awrite, i, msec = 10, retry = 0, retry_max = 20;
 128         struct vmbus_channel *chn;
 129
 130         netif_tx_disable(net);
 131
 132         /* Make sure netvsc_set_multicast_list doesn't re-enable filter! */
 133         cancel_work_sync(&net_device_ctx->work);
 134         ret = rndis_filter_close(device_obj);
 135         if (ret != 0) {
 136                 netdev_err(net, "unable to close device (ret %d).\n", ret);
 137                 return ret;
 138         }
 139
 140         /* Ensure pending bytes in ring are read */
 141         while (true) {
 142                 aread = 0;
 143                 for (i = 0; i < nvdev->num_chn; i++) {
 144                         chn = nvdev->chn_table[i];
 145                         if (!chn)
 146                                 continue;
 147
 148                         hv_get_ringbuffer_availbytes(&chn->inbound, &aread,
 149                                                      &awrite);
 150
 151                         if (aread)
 152                                 break;
 153
 154                         hv_get_ringbuffer_availbytes(&chn->outbound, &aread,
 155                                                      &awrite);
 156
 157                         if (aread)
 158                                 break;
 159                 }
 160
 161                 retry++;
 162                 if (retry > retry_max || aread == 0)
 163                         break;
 164
 165                 msleep(msec);
 166
 167                 if (msec < 1000)
 168                         msec *= 2;
 169         }
 170
 171         if (aread) {
 172                 netdev_err(net, "Ring buffer not empty after closing rndis\n");
 173                 ret = -ETIMEDOUT;
 174         }
 175
 176         return ret;
 177 }
 178
 179 static void *init_ppi_data(struct rndis_message *msg, u32 ppi_size,
 180                                 int pkt_type)
 181 {
 182         struct rndis_packet *rndis_pkt;
 183         struct rndis_per_packet_info *ppi;
 184
 185         rndis_pkt = &msg->msg.pkt;
 186         rndis_pkt->data_offset += ppi_size;
 187
 188         ppi = (struct rndis_per_packet_info *)((void *)rndis_pkt +
 189                 rndis_pkt->per_pkt_info_offset + rndis_pkt->per_pkt_info_len);
 190
 191         ppi->size = ppi_size;
 192         ppi->type = pkt_type;
 193         ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
 194
 195         rndis_pkt->per_pkt_info_len += ppi_size;
 196
 197         return ppi;
 198 }
 199
 200 union sub_key {
 201         u64 k;
 202         struct {
 203                 u8 pad[3];
 204                 u8 kb;
 205                 u32 ka;
 206         };
 207 };
 208
 209 /* Toeplitz hash function
 210  * data: network byte order
 211  * return: host byte order
 212  */
 213 static u32 comp_hash(u8 *key, int klen, void *data, int dlen)
 214 {
 215         union sub_key subk;
 216         int k_next = 4;
 217         u8 dt;
 218         int i, j;
 219         u32 ret = 0;
 220
 221         subk.k = 0;
 222         subk.ka = ntohl(*(u32 *)key);
 223
 224         for (i = 0; i < dlen; i++) {
 225                 subk.kb = key[k_next];
 226                 k_next = (k_next + 1) % klen;
 227                 dt = ((u8 *)data)[i];
 228                 for (j = 0; j < 8; j++) {
 229                         if (dt & 0x80)
 230                                 ret ^= subk.ka;
 231                         dt <<= 1;
 232                         subk.k <<= 1;
 233                 }
 234         }
 235
 236         return ret;
 237 }
 238
 239 static bool netvsc_set_hash(u32 *hash, struct sk_buff *skb)
 240 {
 241         struct flow_keys flow;
 242         int data_len;
 243
 244         if (!skb_flow_dissect_flow_keys(skb, &flow, 0) ||
 245             !(flow.basic.n_proto == htons(ETH_P_IP) ||
 246               flow.basic.n_proto == htons(ETH_P_IPV6)))
 247                 return false;
 248
 249         if (flow.basic.ip_proto == IPPROTO_TCP)
 250                 data_len = 12;
 251         else
 252                 data_len = 8;
 253
 254         *hash = comp_hash(netvsc_hash_key, HASH_KEYLEN, &flow, data_len);
 255
 256         return true;
 257 }
 258
 259 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
 260                         void *accel_priv, select_queue_fallback_t fallback)
 261 {
 262         struct net_device_context *net_device_ctx = netdev_priv(ndev);
 263         struct hv_device *hdev =  net_device_ctx->device_ctx;
 264         struct netvsc_device *nvsc_dev = hv_get_drvdata(hdev);
 265         u32 hash;
 266         u16 q_idx = 0;
 267
 268         if (nvsc_dev == NULL || ndev->real_num_tx_queues <= 1)
 269                 return 0;
 270
 271         if (netvsc_set_hash(&hash, skb)) {
 272                 q_idx = nvsc_dev->send_table[hash % VRSS_SEND_TAB_SIZE] %
 273                         ndev->real_num_tx_queues;
 274                 skb_set_hash(skb, hash, PKT_HASH_TYPE_L3);
 275         }
 276
 277         return q_idx;
 278 }
 279
 280 void netvsc_xmit_completion(void *context)
 281 {
 282         struct hv_netvsc_packet *packet = (struct hv_netvsc_packet *)context;
 283         struct sk_buff *skb = (struct sk_buff *)
 284                 (unsigned long)packet->send_completion_tid;
 285
 286         if (skb)
 287                 dev_kfree_skb_any(skb);
 288 }
 289
 290 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
 291                         struct hv_page_buffer *pb)
 292 {
 293         int j = 0;
 294
 295         /* Deal with compund pages by ignoring unused part
 296          * of the page.
 297          */
 298         page += (offset >> PAGE_SHIFT);
 299         offset &= ~PAGE_MASK;
 300
 301         while (len > 0) {
 302                 unsigned long bytes;
 303
 304                 bytes = PAGE_SIZE - offset;
 305                 if (bytes > len)
 306                         bytes = len;
 307                 pb[j].pfn = page_to_pfn(page);
 308                 pb[j].offset = offset;
 309                 pb[j].len = bytes;
 310
 311                 offset += bytes;
 312                 len -= bytes;
 313
 314                 if (offset == PAGE_SIZE && len) {
 315                         page++;
 316                         offset = 0;
 317                         j++;
 318                 }
 319         }
 320
 321         return j + 1;
 322 }
 323
 324 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
 325                            struct hv_netvsc_packet *packet)
 326 {
 327         struct hv_page_buffer *pb = packet->page_buf;
 328         u32 slots_used = 0;
 329         char *data = skb->data;
 330         int frags = skb_shinfo(skb)->nr_frags;
 331         int i;
 332
 333         /* The packet is laid out thus:
 334          * 1. hdr: RNDIS header and PPI
 335          * 2. skb linear data
 336          * 3. skb fragment data
 337          */
 338         if (hdr != NULL)
 339                 slots_used += fill_pg_buf(virt_to_page(hdr),
 340                                         offset_in_page(hdr),
 341                                         len, &pb[slots_used]);
 342
 343         packet->rmsg_size = len;
 344         packet->rmsg_pgcnt = slots_used;
 345
 346         slots_used += fill_pg_buf(virt_to_page(data),
 347                                 offset_in_page(data),
 348                                 skb_headlen(skb), &pb[slots_used]);
 349
 350         for (i = 0; i < frags; i++) {
 351                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 352
 353                 slots_used += fill_pg_buf(skb_frag_page(frag),
 354                                         frag->page_offset,
 355                                         skb_frag_size(frag), &pb[slots_used]);
 356         }
 357         return slots_used;
 358 }
 359
 360 static int count_skb_frag_slots(struct sk_buff *skb)
 361 {
 362         int i, frags = skb_shinfo(skb)->nr_frags;
 363         int pages = 0;
 364
 365         for (i = 0; i < frags; i++) {
 366                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 367                 unsigned long size = skb_frag_size(frag);
 368                 unsigned long offset = frag->page_offset;
 369
 370                 /* Skip unused frames from start of page */
 371                 offset &= ~PAGE_MASK;
 372                 pages += PFN_UP(offset + size);
 373         }
 374         return pages;
 375 }
 376
 377 static int netvsc_get_slots(struct sk_buff *skb)
 378 {
 379         char *data = skb->data;
 380         unsigned int offset = offset_in_page(data);
 381         unsigned int len = skb_headlen(skb);
 382         int slots;
 383         int frag_slots;
 384
 385         slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
 386         frag_slots = count_skb_frag_slots(skb);
 387         return slots + frag_slots;
 388 }
 389
 390 static u32 get_net_transport_info(struct sk_buff *skb, u32 *trans_off)
 391 {
 392         u32 ret_val = TRANSPORT_INFO_NOT_IP;
 393
 394         if ((eth_hdr(skb)->h_proto != htons(ETH_P_IP)) &&
 395                 (eth_hdr(skb)->h_proto != htons(ETH_P_IPV6))) {
 396                 goto not_ip;
 397         }
 398
 399         *trans_off = skb_transport_offset(skb);
 400
 401         if ((eth_hdr(skb)->h_proto == htons(ETH_P_IP))) {
 402                 struct iphdr *iphdr = ip_hdr(skb);
 403
 404                 if (iphdr->protocol == IPPROTO_TCP)
 405                         ret_val = TRANSPORT_INFO_IPV4_TCP;
 406                 else if (iphdr->protocol == IPPROTO_UDP)
 407                         ret_val = TRANSPORT_INFO_IPV4_UDP;
 408         } else {
 409                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
 410                         ret_val = TRANSPORT_INFO_IPV6_TCP;
 411                 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
 412                         ret_val = TRANSPORT_INFO_IPV6_UDP;
 413         }
 414
 415 not_ip:
 416         return ret_val;
 417 }
 418
 419 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
 420 {
 421         struct net_device_context *net_device_ctx = netdev_priv(net);
 422         struct hv_netvsc_packet *packet = NULL;
 423         int ret;
 424         unsigned int num_data_pgs;
 425         struct rndis_message *rndis_msg;
 426         struct rndis_packet *rndis_pkt;
 427         u32 rndis_msg_size;
 428         bool isvlan;
 429         bool linear = false;
 430         struct rndis_per_packet_info *ppi;
 431         struct ndis_tcp_ip_checksum_info *csum_info;
 432         struct ndis_tcp_lso_info *lso_info;
 433         int  hdr_offset;
 434         u32 net_trans_info;
 435         u32 hash;
 436         u32 skb_length;
 437         u32 pkt_sz;
 438         struct hv_page_buffer page_buf[MAX_PAGE_BUFFER_COUNT];
 439         struct netvsc_stats *tx_stats = this_cpu_ptr(net_device_ctx->tx_stats);
 440
 441         /* We will atmost need two pages to describe the rndis
 442          * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
 443          * of pages in a single packet. If skb is scattered around
 444          * more pages we try linearizing it.
 445          */
 446
 447 check_size:
 448         skb_length = skb->len;
 449         num_data_pgs = netvsc_get_slots(skb) + 2;
 450         if (num_data_pgs > MAX_PAGE_BUFFER_COUNT && linear) {
 451                 net_alert_ratelimited("packet too big: %u pages (%u bytes)\n",
 452                                       num_data_pgs, skb->len);
 453                 ret = -EFAULT;
 454                 goto drop;
 455         } else if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
 456                 if (skb_linearize(skb)) {
 457                         net_alert_ratelimited("failed to linearize skb\n");
 458                         ret = -ENOMEM;
 459                         goto drop;
 460                 }
 461                 linear = true;
 462                 goto check_size;
 463         }
 464
 465         pkt_sz = sizeof(struct hv_netvsc_packet) + RNDIS_AND_PPI_SIZE;
 466
 467         ret = skb_cow_head(skb, pkt_sz);
 468         if (ret) {
 469                 netdev_err(net, "unable to alloc hv_netvsc_packet\n");
 470                 ret = -ENOMEM;
 471                 goto drop;
 472         }
 473         /* Use the headroom for building up the packet */
 474         packet = (struct hv_netvsc_packet *)skb->head;
 475
 476         packet->status = 0;
 477         packet->xmit_more = skb->xmit_more;
 478
 479         packet->vlan_tci = skb->vlan_tci;
 480         packet->page_buf = page_buf;
 481
 482         packet->q_idx = skb_get_queue_mapping(skb);
 483
 484         packet->is_data_pkt = true;
 485         packet->total_data_buflen = skb->len;
 486
 487         packet->rndis_msg = (struct rndis_message *)((unsigned long)packet +
 488                                 sizeof(struct hv_netvsc_packet));
 489
 490         memset(packet->rndis_msg, 0, RNDIS_AND_PPI_SIZE);
 491
 492         /* Set the completion routine */
 493         packet->send_completion = netvsc_xmit_completion;
 494         packet->send_completion_ctx = packet;
 495         packet->send_completion_tid = (unsigned long)skb;
 496
 497         isvlan = packet->vlan_tci & VLAN_TAG_PRESENT;
 498
 499         /* Add the rndis header */
 500         rndis_msg = packet->rndis_msg;
 501         rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
 502         rndis_msg->msg_len = packet->total_data_buflen;
 503         rndis_pkt = &rndis_msg->msg.pkt;
 504         rndis_pkt->data_offset = sizeof(struct rndis_packet);
 505         rndis_pkt->data_len = packet->total_data_buflen;
 506         rndis_pkt->per_pkt_info_offset = sizeof(struct rndis_packet);
 507
 508         rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
 509
 510         hash = skb_get_hash_raw(skb);
 511         if (hash != 0 && net->real_num_tx_queues > 1) {
 512                 rndis_msg_size += NDIS_HASH_PPI_SIZE;
 513                 ppi = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
 514                                     NBL_HASH_VALUE);
 515                 *(u32 *)((void *)ppi + ppi->ppi_offset) = hash;
 516         }
 517
 518         if (isvlan) {
 519                 struct ndis_pkt_8021q_info *vlan;
 520
 521                 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
 522                 ppi = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
 523                                         IEEE_8021Q_INFO);
 524                 vlan = (struct ndis_pkt_8021q_info *)((void *)ppi +
 525                                                 ppi->ppi_offset);
 526                 vlan->vlanid = packet->vlan_tci & VLAN_VID_MASK;
 527                 vlan->pri = (packet->vlan_tci & VLAN_PRIO_MASK) >>
 528                                 VLAN_PRIO_SHIFT;
 529         }
 530
 531         net_trans_info = get_net_transport_info(skb, &hdr_offset);
 532         if (net_trans_info == TRANSPORT_INFO_NOT_IP)
 533                 goto do_send;
 534
 535         /*
 536          * Setup the sendside checksum offload only if this is not a
 537          * GSO packet.
 538          */
 539         if (skb_is_gso(skb))
 540                 goto do_lso;
 541
 542         if ((skb->ip_summed == CHECKSUM_NONE) ||
 543             (skb->ip_summed == CHECKSUM_UNNECESSARY))
 544                 goto do_send;
 545
 546         rndis_msg_size += NDIS_CSUM_PPI_SIZE;
 547         ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
 548                             TCPIP_CHKSUM_PKTINFO);
 549
 550         csum_info = (struct ndis_tcp_ip_checksum_info *)((void *)ppi +
 551                         ppi->ppi_offset);
 552
 553         if (net_trans_info & (INFO_IPV4 << 16))
 554                 csum_info->transmit.is_ipv4 = 1;
 555         else
 556                 csum_info->transmit.is_ipv6 = 1;
 557
 558         if (net_trans_info & INFO_TCP) {
 559                 csum_info->transmit.tcp_checksum = 1;
 560                 csum_info->transmit.tcp_header_offset = hdr_offset;
 561         } else if (net_trans_info & INFO_UDP) {
 562                 /* UDP checksum offload is not supported on ws2008r2.
 563                  * Furthermore, on ws2012 and ws2012r2, there are some
 564                  * issues with udp checksum offload from Linux guests.
 565                  * (these are host issues).
 566                  * For now compute the checksum here.
 567                  */
 568                 struct udphdr *uh;
 569                 u16 udp_len;
 570
 571                 ret = skb_cow_head(skb, 0);
 572                 if (ret)
 573                         goto drop;
 574
 575                 uh = udp_hdr(skb);
 576                 udp_len = ntohs(uh->len);
 577                 uh->check = 0;
 578                 uh->check = csum_tcpudp_magic(ip_hdr(skb)->saddr,
 579                                               ip_hdr(skb)->daddr,
 580                                               udp_len, IPPROTO_UDP,
 581                                               csum_partial(uh, udp_len, 0));
 582                 if (uh->check == 0)
 583                         uh->check = CSUM_MANGLED_0;
 584
 585                 csum_info->transmit.udp_checksum = 0;
 586         }
 587         goto do_send;
 588
 589 do_lso:
 590         rndis_msg_size += NDIS_LSO_PPI_SIZE;
 591         ppi = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
 592                             TCP_LARGESEND_PKTINFO);
 593
 594         lso_info = (struct ndis_tcp_lso_info *)((void *)ppi +
 595                         ppi->ppi_offset);
 596
 597         lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
 598         if (net_trans_info & (INFO_IPV4 << 16)) {
 599                 lso_info->lso_v2_transmit.ip_version =
 600                         NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
 601                 ip_hdr(skb)->tot_len = 0;
 602                 ip_hdr(skb)->check = 0;
 603                 tcp_hdr(skb)->check =
 604                 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
 605                                    ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 606         } else {
 607                 lso_info->lso_v2_transmit.ip_version =
 608                         NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
 609                 ipv6_hdr(skb)->payload_len = 0;
 610                 tcp_hdr(skb)->check =
 611                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 612                                 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 613         }
 614         lso_info->lso_v2_transmit.tcp_header_offset = hdr_offset;
 615         lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
 616
 617 do_send:
 618         /* Start filling in the page buffers with the rndis hdr */
 619         rndis_msg->msg_len += rndis_msg_size;
 620         packet->total_data_buflen = rndis_msg->msg_len;
 621         packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
 622                                                skb, packet);
 623
 624         ret = netvsc_send(net_device_ctx->device_ctx, packet);
 625
 626 drop:
 627         if (ret == 0) {
 628                 u64_stats_update_begin(&tx_stats->syncp);
 629                 tx_stats->packets++;
 630                 tx_stats->bytes += skb_length;
 631                 u64_stats_update_end(&tx_stats->syncp);
 632         } else {
 633                 if (ret != -EAGAIN) {
 634                         dev_kfree_skb_any(skb);
 635                         net->stats.tx_dropped++;
 636                 }
 637         }
 638
 639         return (ret == -EAGAIN) ? NETDEV_TX_BUSY : NETDEV_TX_OK;
 640 }
 641
 642 /*
 643  * netvsc_linkstatus_callback - Link up/down notification
 644  */
 645 void netvsc_linkstatus_callback(struct hv_device *device_obj,
 646                                 struct rndis_message *resp)
 647 {
 648         struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
 649         struct net_device *net;
 650         struct net_device_context *ndev_ctx;
 651         struct netvsc_device *net_device;
 652         struct rndis_device *rdev;
 653
 654         net_device = hv_get_drvdata(device_obj);
 655         rdev = net_device->extension;
 656
 657         switch (indicate->status) {
 658         case RNDIS_STATUS_MEDIA_CONNECT:
 659                 rdev->link_state = false;
 660                 break;
 661         case RNDIS_STATUS_MEDIA_DISCONNECT:
 662                 rdev->link_state = true;
 663                 break;
 664         case RNDIS_STATUS_NETWORK_CHANGE:
 665                 rdev->link_change = true;
 666                 break;
 667         default:
 668                 return;
 669         }
 670
 671         net = net_device->ndev;
 672
 673         if (!net || net->reg_state != NETREG_REGISTERED)
 674                 return;
 675
 676         ndev_ctx = netdev_priv(net);
 677         if (!rdev->link_state) {
 678                 schedule_delayed_work(&ndev_ctx->dwork, 0);
 679                 schedule_delayed_work(&ndev_ctx->dwork, msecs_to_jiffies(20));
 680         } else {
 681                 schedule_delayed_work(&ndev_ctx->dwork, 0);
 682         }
 683 }
 684
 685 /*
 686  * netvsc_recv_callback -  Callback when we receive a packet from the
 687  * "wire" on the specified device.
 688  */
 689 int netvsc_recv_callback(struct hv_device *device_obj,
 690                                 struct hv_netvsc_packet *packet,
 691                                 struct ndis_tcp_ip_checksum_info *csum_info)
 692 {
 693         struct net_device *net;
 694         struct net_device_context *net_device_ctx;
 695         struct sk_buff *skb;
 696         struct netvsc_stats *rx_stats;
 697
 698         net = ((struct netvsc_device *)hv_get_drvdata(device_obj))->ndev;
 699         if (!net || net->reg_state != NETREG_REGISTERED) {
 700                 packet->status = NVSP_STAT_FAIL;
 701                 return 0;
 702         }
 703         net_device_ctx = netdev_priv(net);
 704         rx_stats = this_cpu_ptr(net_device_ctx->rx_stats);
 705
 706         /* Allocate a skb - TODO direct I/O to pages? */
 707         skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
 708         if (unlikely(!skb)) {
 709                 ++net->stats.rx_dropped;
 710                 packet->status = NVSP_STAT_FAIL;
 711                 return 0;
 712         }
 713
 714         /*
 715          * Copy to skb. This copy is needed here since the memory pointed by
 716          * hv_netvsc_packet cannot be deallocated
 717          */
 718         memcpy(skb_put(skb, packet->total_data_buflen), packet->data,
 719                 packet->total_data_buflen);
 720
 721         skb->protocol = eth_type_trans(skb, net);
 722         if (csum_info) {
 723                 /* We only look at the IP checksum here.
 724                  * Should we be dropping the packet if checksum
 725                  * failed? How do we deal with other checksums - TCP/UDP?
 726                  */
 727                 if (csum_info->receive.ip_checksum_succeeded)
 728                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 729                 else
 730                         skb->ip_summed = CHECKSUM_NONE;
 731         }
 732
 733         if (packet->vlan_tci & VLAN_TAG_PRESENT)
 734                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 735                                        packet->vlan_tci);
 736
 737         skb_record_rx_queue(skb, packet->channel->
 738                             offermsg.offer.sub_channel_index);
 739
 740         u64_stats_update_begin(&rx_stats->syncp);
 741         rx_stats->packets++;
 742         rx_stats->bytes += packet->total_data_buflen;
 743         u64_stats_update_end(&rx_stats->syncp);
 744
 745         /*
 746          * Pass the skb back up. Network stack will deallocate the skb when it
 747          * is done.
 748          * TODO - use NAPI?
 749          */
 750         netif_rx(skb);
 751
 752         return 0;
 753 }
 754
 755 static void netvsc_get_drvinfo(struct net_device *net,
 756                                struct ethtool_drvinfo *info)
 757 {
 758         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
 759         strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
 760 }
 761
 762 static void netvsc_get_channels(struct net_device *net,
 763                                 struct ethtool_channels *channel)
 764 {
 765         struct net_device_context *net_device_ctx = netdev_priv(net);
 766         struct hv_device *dev = net_device_ctx->device_ctx;
 767         struct netvsc_device *nvdev = hv_get_drvdata(dev);
 768
 769         if (nvdev) {
 770                 channel->max_combined   = nvdev->max_chn;
 771                 channel->combined_count = nvdev->num_chn;
 772         }
 773 }
 774
 775 static int netvsc_set_channels(struct net_device *net,
 776                                struct ethtool_channels *channels)
 777 {
 778         struct net_device_context *net_device_ctx = netdev_priv(net);
 779         struct hv_device *dev = net_device_ctx->device_ctx;
 780         struct netvsc_device *nvdev = hv_get_drvdata(dev);
 781         struct netvsc_device_info device_info;
 782         u32 num_chn;
 783         u32 max_chn;
 784         int ret = 0;
 785         bool recovering = false;
 786
 787         if (!nvdev || nvdev->destroy)
 788                 return -ENODEV;
 789
 790         num_chn = nvdev->num_chn;
 791         max_chn = min_t(u32, nvdev->max_chn, num_online_cpus());
 792
 793         if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5) {
 794                 pr_info("vRSS unsupported before NVSP Version 5\n");
 795                 return -EINVAL;
 796         }
 797
 798         /* We do not support rx, tx, or other */
 799         if (!channels ||
 800             channels->rx_count ||
 801             channels->tx_count ||
 802             channels->other_count ||
 803             (channels->combined_count < 1))
 804                 return -EINVAL;
 805
 806         if (channels->combined_count > max_chn) {
 807                 pr_info("combined channels too high, using %d\n", max_chn);
 808                 channels->combined_count = max_chn;
 809         }
 810
 811         ret = netvsc_close(net);
 812         if (ret)
 813                 goto out;
 814
 815  do_set:
 816         nvdev->start_remove = true;
 817         rndis_filter_device_remove(dev);
 818
 819         nvdev->num_chn = channels->combined_count;
 820
 821         net_device_ctx->device_ctx = dev;
 822         hv_set_drvdata(dev, net);
 823
 824         memset(&device_info, 0, sizeof(device_info));
 825         device_info.num_chn = nvdev->num_chn; /* passed to RNDIS */
 826         device_info.ring_size = ring_size;
 827         device_info.max_num_vrss_chns = max_num_vrss_chns;
 828
 829         ret = rndis_filter_device_add(dev, &device_info);
 830         if (ret) {
 831                 if (recovering) {
 832                         netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
 833                         return ret;
 834                 }
 835                 goto recover;
 836         }
 837
 838         nvdev = hv_get_drvdata(dev);
 839
 840         ret = netif_set_real_num_tx_queues(net, nvdev->num_chn);
 841         if (ret) {
 842                 if (recovering) {
 843                         netdev_err(net, "could not set tx queue count (ret %d)\n", ret);
 844                         return ret;
 845                 }
 846                 goto recover;
 847         }
 848
 849         ret = netif_set_real_num_rx_queues(net, nvdev->num_chn);
 850         if (ret) {
 851                 if (recovering) {
 852                         netdev_err(net, "could not set rx queue count (ret %d)\n", ret);
 853                         return ret;
 854                 }
 855                 goto recover;
 856         }
 857         netif_set_gso_max_size(net, NETVSC_GSO_MAX_SIZE);
 858
 859  out:
 860         netvsc_open(net);
 861
 862         return ret;
 863
 864  recover:
 865         /* If the above failed, we attempt to recover through the same
 866          * process but with the original number of channels.
 867          */
 868         netdev_err(net, "could not set channels, recovering\n");
 869         recovering = true;
 870         channels->combined_count = num_chn;
 871         goto do_set;
 872 }
 873
 874 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
 875 {
 876         struct net_device_context *ndevctx = netdev_priv(ndev);
 877         struct hv_device *hdev =  ndevctx->device_ctx;
 878         struct netvsc_device *nvdev = hv_get_drvdata(hdev);
 879         struct netvsc_device_info device_info;
 880         int limit = ETH_DATA_LEN;
 881         int ret = 0;
 882
 883         if (nvdev == NULL || nvdev->destroy)
 884                 return -ENODEV;
 885
 886         if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
 887                 limit = NETVSC_MTU - ETH_HLEN;
 888
 889         if (mtu < NETVSC_MTU_MIN || mtu > limit)
 890                 return -EINVAL;
 891
 892         ret = netvsc_close(ndev);
 893         if (ret)
 894                 goto out;
 895
 896         nvdev->start_remove = true;
 897         rndis_filter_device_remove(hdev);
 898
 899         ndev->mtu = mtu;
 900
 901         ndevctx->device_ctx = hdev;
 902         hv_set_drvdata(hdev, ndev);
 903
 904         memset(&device_info, 0, sizeof(device_info));
 905         device_info.ring_size = ring_size;
 906         device_info.num_chn = nvdev->num_chn;
 907         device_info.max_num_vrss_chns = max_num_vrss_chns;
 908         rndis_filter_device_add(hdev, &device_info);
 909
 910 out:
 911         netvsc_open(ndev);
 912
 913         return ret;
 914 }
 915
 916 static struct rtnl_link_stats64 *netvsc_get_stats64(struct net_device *net,
 917                                                     struct rtnl_link_stats64 *t)
 918 {
 919         struct net_device_context *ndev_ctx = netdev_priv(net);
 920         int cpu;
 921
 922         for_each_possible_cpu(cpu) {
 923                 struct netvsc_stats *tx_stats = per_cpu_ptr(ndev_ctx->tx_stats,
 924                                                             cpu);
 925                 struct netvsc_stats *rx_stats = per_cpu_ptr(ndev_ctx->rx_stats,
 926                                                             cpu);
 927                 u64 tx_packets, tx_bytes, rx_packets, rx_bytes;
 928                 unsigned int start;
 929
 930                 do {
 931                         start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
 932                         tx_packets = tx_stats->packets;
 933                         tx_bytes = tx_stats->bytes;
 934                 } while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
 935
 936                 do {
 937                         start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
 938                         rx_packets = rx_stats->packets;
 939                         rx_bytes = rx_stats->bytes;
 940                 } while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
 941
 942                 t->tx_bytes     += tx_bytes;
 943                 t->tx_packets   += tx_packets;
 944                 t->rx_bytes     += rx_bytes;
 945                 t->rx_packets   += rx_packets;
 946         }
 947
 948         t->tx_dropped   = net->stats.tx_dropped;
 949         t->tx_errors    = net->stats.tx_dropped;
 950
 951         t->rx_dropped   = net->stats.rx_dropped;
 952         t->rx_errors    = net->stats.rx_errors;
 953
 954         return t;
 955 }
 956
 957 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
 958 {
 959         struct net_device_context *ndevctx = netdev_priv(ndev);
 960         struct hv_device *hdev =  ndevctx->device_ctx;
 961         struct sockaddr *addr = p;
 962         char save_adr[ETH_ALEN];
 963         unsigned char save_aatype;
 964         int err;
 965
 966         memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
 967         save_aatype = ndev->addr_assign_type;
 968
 969         err = eth_mac_addr(ndev, p);
 970         if (err != 0)
 971                 return err;
 972
 973         err = rndis_filter_set_device_mac(hdev, addr->sa_data);
 974         if (err != 0) {
 975                 /* roll back to saved MAC */
 976                 memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
 977                 ndev->addr_assign_type = save_aatype;
 978         }
 979
 980         return err;
 981 }
 982
 983 #ifdef CONFIG_NET_POLL_CONTROLLER
 984 static void netvsc_poll_controller(struct net_device *net)
 985 {
 986         /* As netvsc_start_xmit() works synchronous we don't have to
 987          * trigger anything here.
 988          */
 989 }
 990 #endif
 991
 992 static const struct ethtool_ops ethtool_ops = {
 993         .get_drvinfo    = netvsc_get_drvinfo,
 994         .get_link       = ethtool_op_get_link,
 995         .get_channels   = netvsc_get_channels,
 996         .set_channels   = netvsc_set_channels,
 997 };
 998
 999 static const struct net_device_ops device_ops = {
1000         .ndo_open =                     netvsc_open,
1001         .ndo_stop =                     netvsc_close,
1002         .ndo_start_xmit =               netvsc_start_xmit,
1003         .ndo_set_rx_mode =              netvsc_set_multicast_list,
1004         .ndo_change_mtu =               netvsc_change_mtu,
1005         .ndo_validate_addr =            eth_validate_addr,
1006         .ndo_set_mac_address =          netvsc_set_mac_addr,
1007         .ndo_select_queue =             netvsc_select_queue,
1008         .ndo_get_stats64 =              netvsc_get_stats64,
1009 #ifdef CONFIG_NET_POLL_CONTROLLER
1010         .ndo_poll_controller =          netvsc_poll_controller,
1011 #endif
1012 };
1013
1014 /*
1015  * Send GARP packet to network peers after migrations.
1016  * After Quick Migration, the network is not immediately operational in the
1017  * current context when receiving RNDIS_STATUS_MEDIA_CONNECT event. So, add
1018  * another netif_notify_peers() into a delayed work, otherwise GARP packet
1019  * will not be sent after quick migration, and cause network disconnection.
1020  * Also, we update the carrier status here.
1021  */
1022 static void netvsc_link_change(struct work_struct *w)
1023 {
1024         struct net_device_context *ndev_ctx;
1025         struct net_device *net;
1026         struct netvsc_device *net_device;
1027         struct rndis_device *rdev;
1028         bool notify, refresh = false;
1029         char *argv[] = { "/etc/init.d/network", "restart", NULL };
1030         char *envp[] = { "HOME=/", "PATH=/sbin:/usr/sbin:/bin:/usr/bin", NULL };
1031
1032         rtnl_lock();
1033
1034         ndev_ctx = container_of(w, struct net_device_context, dwork.work);
1035         net_device = hv_get_drvdata(ndev_ctx->device_ctx);
1036         rdev = net_device->extension;
1037         net = net_device->ndev;
1038
1039         if (rdev->link_state) {
1040                 netif_carrier_off(net);
1041                 notify = false;
1042         } else {
1043                 netif_carrier_on(net);
1044                 notify = true;
1045                 if (rdev->link_change) {
1046                         rdev->link_change = false;
1047                         refresh = true;
1048                 }
1049         }
1050
1051         rtnl_unlock();
1052
1053         if (refresh)
1054                 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
1055
1056         if (notify)
1057                 netdev_notify_peers(net);
1058 }
1059
1060 static void netvsc_free_netdev(struct net_device *netdev)
1061 {
1062         struct net_device_context *net_device_ctx = netdev_priv(netdev);
1063
1064         free_percpu(net_device_ctx->tx_stats);
1065         free_percpu(net_device_ctx->rx_stats);
1066         free_netdev(netdev);
1067 }
1068
1069 static int netvsc_probe(struct hv_device *dev,
1070                         const struct hv_vmbus_device_id *dev_id)
1071 {
1072         struct net_device *net = NULL;
1073         struct net_device_context *net_device_ctx;
1074         struct netvsc_device_info device_info;
1075         struct netvsc_device *nvdev;
1076         int ret;
1077         u32 max_needed_headroom;
1078
1079         net = alloc_etherdev_mq(sizeof(struct net_device_context),
1080                                 num_online_cpus());
1081         if (!net)
1082                 return -ENOMEM;
1083
1084         max_needed_headroom = sizeof(struct hv_netvsc_packet) +
1085                               RNDIS_AND_PPI_SIZE;
1086
1087         netif_carrier_off(net);
1088
1089         net_device_ctx = netdev_priv(net);
1090         net_device_ctx->device_ctx = dev;
1091         net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
1092         if (netif_msg_probe(net_device_ctx))
1093                 netdev_dbg(net, "netvsc msg_enable: %d\n",
1094                            net_device_ctx->msg_enable);
1095
1096         net_device_ctx->tx_stats = netdev_alloc_pcpu_stats(struct netvsc_stats);
1097         if (!net_device_ctx->tx_stats) {
1098                 free_netdev(net);
1099                 return -ENOMEM;
1100         }
1101         net_device_ctx->rx_stats = netdev_alloc_pcpu_stats(struct netvsc_stats);
1102         if (!net_device_ctx->rx_stats) {
1103                 free_percpu(net_device_ctx->tx_stats);
1104                 free_netdev(net);
1105                 return -ENOMEM;
1106         }
1107
1108         hv_set_drvdata(dev, net);
1109         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
1110         INIT_WORK(&net_device_ctx->work, do_set_multicast);
1111
1112         net->netdev_ops = &device_ops;
1113
1114         net->hw_features = NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_IP_CSUM |
1115                                 NETIF_F_TSO;
1116         net->features = NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_SG | NETIF_F_RXCSUM |
1117                         NETIF_F_IP_CSUM | NETIF_F_TSO;
1118
1119         net->ethtool_ops = &ethtool_ops;
1120         SET_NETDEV_DEV(net, &dev->device);
1121
1122         /*
1123          * Request additional head room in the skb.
1124          * We will use this space to build the rndis
1125          * heaser and other state we need to maintain.
1126          */
1127         net->needed_headroom = max_needed_headroom;
1128
1129         /* Notify the netvsc driver of the new device */
1130         memset(&device_info, 0, sizeof(device_info));
1131         device_info.ring_size = ring_size;
1132         device_info.max_num_vrss_chns = max_num_vrss_chns;
1133         ret = rndis_filter_device_add(dev, &device_info);
1134         if (ret != 0) {
1135                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
1136                 netvsc_free_netdev(net);
1137                 hv_set_drvdata(dev, NULL);
1138                 return ret;
1139         }
1140         memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
1141
1142         nvdev = hv_get_drvdata(dev);
1143         netif_set_real_num_tx_queues(net, nvdev->num_chn);
1144         netif_set_real_num_rx_queues(net, nvdev->num_chn);
1145
1146         ret = register_netdev(net);
1147         if (ret != 0) {
1148                 pr_err("Unable to register netdev.\n");
1149                 rndis_filter_device_remove(dev);
1150                 netvsc_free_netdev(net);
1151         } else {
1152                 schedule_delayed_work(&net_device_ctx->dwork, 0);
1153         }
1154
1155         return ret;
1156 }
1157
1158 static int netvsc_remove(struct hv_device *dev)
1159 {
1160         struct net_device *net;
1161         struct net_device_context *ndev_ctx;
1162         struct netvsc_device *net_device;
1163
1164         net_device = hv_get_drvdata(dev);
1165         net = net_device->ndev;
1166
1167         if (net == NULL) {
1168                 dev_err(&dev->device, "No net device to remove\n");
1169                 return 0;
1170         }
1171
1172         net_device->start_remove = true;
1173
1174         ndev_ctx = netdev_priv(net);
1175         cancel_delayed_work_sync(&ndev_ctx->dwork);
1176         cancel_work_sync(&ndev_ctx->work);
1177
1178         /* Stop outbound asap */
1179         netif_tx_disable(net);
1180
1181         unregister_netdev(net);
1182
1183         /*
1184          * Call to the vsc driver to let it know that the device is being
1185          * removed
1186          */
1187         rndis_filter_device_remove(dev);
1188
1189         netvsc_free_netdev(net);
1190         return 0;
1191 }
1192
1193 static const struct hv_vmbus_device_id id_table[] = {
1194         /* Network guid */
1195         { HV_NIC_GUID, },
1196         { },
1197 };
1198
1199 MODULE_DEVICE_TABLE(vmbus, id_table);
1200
1201 /* The one and only one */
1202 static struct  hv_driver netvsc_drv = {
1203         .name = KBUILD_MODNAME,
1204         .id_table = id_table,
1205         .probe = netvsc_probe,
1206         .remove = netvsc_remove,
1207 };
1208
1209 static void __exit netvsc_drv_exit(void)
1210 {
1211         vmbus_driver_unregister(&netvsc_drv);
1212 }
1213
1214 static int __init netvsc_drv_init(void)
1215 {
1216         if (ring_size < RING_SIZE_MIN) {
1217                 ring_size = RING_SIZE_MIN;
1218                 pr_info("Increased ring_size to %d (min allowed)\n",
1219                         ring_size);
1220         }
1221         return vmbus_driver_register(&netvsc_drv);
1222 }
1223
1224 MODULE_LICENSE("GPL");
1225 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
1226
1227 module_init(netvsc_drv_init);
1228 module_exit(netvsc_drv_exit);