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