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 #define RING_SIZE_MIN 64
  44 #define LINKCHANGE_INT (2 * HZ)
  45
  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 const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
  51                                 NETIF_MSG_LINK | NETIF_MSG_IFUP |
  52                                 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
  53                                 NETIF_MSG_TX_ERR;
  54
  55 static int debug = -1;
  56 module_param(debug, int, S_IRUGO);
  57 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  58
  59 static void netvsc_set_multicast_list(struct net_device *net)
  60 {
  61         struct net_device_context *net_device_ctx = netdev_priv(net);
  62         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
  63
  64         rndis_filter_update(nvdev);
  65 }
  66
  67 static int netvsc_open(struct net_device *net)
  68 {
  69         struct netvsc_device *nvdev = net_device_to_netvsc_device(net);
  70         struct rndis_device *rdev;
  71         int ret = 0;
  72
  73         netif_carrier_off(net);
  74
  75         /* Open up the device */
  76         ret = rndis_filter_open(nvdev);
  77         if (ret != 0) {
  78                 netdev_err(net, "unable to open device (ret %d).\n", ret);
  79                 return ret;
  80         }
  81
  82         netif_tx_wake_all_queues(net);
  83
  84         rdev = nvdev->extension;
  85         if (!rdev->link_state)
  86                 netif_carrier_on(net);
  87
  88         return ret;
  89 }
  90
  91 static int netvsc_close(struct net_device *net)
  92 {
  93         struct net_device_context *net_device_ctx = netdev_priv(net);
  94         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
  95         int ret;
  96         u32 aread, awrite, i, msec = 10, retry = 0, retry_max = 20;
  97         struct vmbus_channel *chn;
  98
  99         netif_tx_disable(net);
 100
 101         ret = rndis_filter_close(nvdev);
 102         if (ret != 0) {
 103                 netdev_err(net, "unable to close device (ret %d).\n", ret);
 104                 return ret;
 105         }
 106
 107         /* Ensure pending bytes in ring are read */
 108         while (true) {
 109                 aread = 0;
 110                 for (i = 0; i < nvdev->num_chn; i++) {
 111                         chn = nvdev->chan_table[i].channel;
 112                         if (!chn)
 113                                 continue;
 114
 115                         hv_get_ringbuffer_availbytes(&chn->inbound, &aread,
 116                                                      &awrite);
 117
 118                         if (aread)
 119                                 break;
 120
 121                         hv_get_ringbuffer_availbytes(&chn->outbound, &aread,
 122                                                      &awrite);
 123
 124                         if (aread)
 125                                 break;
 126                 }
 127
 128                 retry++;
 129                 if (retry > retry_max || aread == 0)
 130                         break;
 131
 132                 msleep(msec);
 133
 134                 if (msec < 1000)
 135                         msec *= 2;
 136         }
 137
 138         if (aread) {
 139                 netdev_err(net, "Ring buffer not empty after closing rndis\n");
 140                 ret = -ETIMEDOUT;
 141         }
 142
 143         return ret;
 144 }
 145
 146 static void *init_ppi_data(struct rndis_message *msg, u32 ppi_size,
 147                                 int pkt_type)
 148 {
 149         struct rndis_packet *rndis_pkt;
 150         struct rndis_per_packet_info *ppi;
 151
 152         rndis_pkt = &msg->msg.pkt;
 153         rndis_pkt->data_offset += ppi_size;
 154
 155         ppi = (struct rndis_per_packet_info *)((void *)rndis_pkt +
 156                 rndis_pkt->per_pkt_info_offset + rndis_pkt->per_pkt_info_len);
 157
 158         ppi->size = ppi_size;
 159         ppi->type = pkt_type;
 160         ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
 161
 162         rndis_pkt->per_pkt_info_len += ppi_size;
 163
 164         return ppi;
 165 }
 166
 167 /* Azure hosts don't support non-TCP port numbers in hashing yet. We compute
 168  * hash for non-TCP traffic with only IP numbers.
 169  */
 170 static inline u32 netvsc_get_hash(struct sk_buff *skb, struct sock *sk)
 171 {
 172         struct flow_keys flow;
 173         u32 hash;
 174         static u32 hashrnd __read_mostly;
 175
 176         net_get_random_once(&hashrnd, sizeof(hashrnd));
 177
 178         if (!skb_flow_dissect_flow_keys(skb, &flow, 0))
 179                 return 0;
 180
 181         if (flow.basic.ip_proto == IPPROTO_TCP) {
 182                 return skb_get_hash(skb);
 183         } else {
 184                 if (flow.basic.n_proto == htons(ETH_P_IP))
 185                         hash = jhash2((u32 *)&flow.addrs.v4addrs, 2, hashrnd);
 186                 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
 187                         hash = jhash2((u32 *)&flow.addrs.v6addrs, 8, hashrnd);
 188                 else
 189                         hash = 0;
 190
 191                 skb_set_hash(skb, hash, PKT_HASH_TYPE_L3);
 192         }
 193
 194         return hash;
 195 }
 196
 197 static inline int netvsc_get_tx_queue(struct net_device *ndev,
 198                                       struct sk_buff *skb, int old_idx)
 199 {
 200         const struct net_device_context *ndc = netdev_priv(ndev);
 201         struct sock *sk = skb->sk;
 202         int q_idx;
 203
 204         q_idx = ndc->tx_send_table[netvsc_get_hash(skb, sk) &
 205                                    (VRSS_SEND_TAB_SIZE - 1)];
 206
 207         /* If queue index changed record the new value */
 208         if (q_idx != old_idx &&
 209             sk && sk_fullsock(sk) && rcu_access_pointer(sk->sk_dst_cache))
 210                 sk_tx_queue_set(sk, q_idx);
 211
 212         return q_idx;
 213 }
 214
 215 /*
 216  * Select queue for transmit.
 217  *
 218  * If a valid queue has already been assigned, then use that.
 219  * Otherwise compute tx queue based on hash and the send table.
 220  *
 221  * This is basically similar to default (__netdev_pick_tx) with the added step
 222  * of using the host send_table when no other queue has been assigned.
 223  *
 224  * TODO support XPS - but get_xps_queue not exported
 225  */
 226 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
 227                         void *accel_priv, select_queue_fallback_t fallback)
 228 {
 229         unsigned int num_tx_queues = ndev->real_num_tx_queues;
 230         int q_idx = sk_tx_queue_get(skb->sk);
 231
 232         if (q_idx < 0 || skb->ooo_okay) {
 233                 /* If forwarding a packet, we use the recorded queue when
 234                  * available for better cache locality.
 235                  */
 236                 if (skb_rx_queue_recorded(skb))
 237                         q_idx = skb_get_rx_queue(skb);
 238                 else
 239                         q_idx = netvsc_get_tx_queue(ndev, skb, q_idx);
 240         }
 241
 242         while (unlikely(q_idx >= num_tx_queues))
 243                 q_idx -= num_tx_queues;
 244
 245         return q_idx;
 246 }
 247
 248 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
 249                         struct hv_page_buffer *pb)
 250 {
 251         int j = 0;
 252
 253         /* Deal with compund pages by ignoring unused part
 254          * of the page.
 255          */
 256         page += (offset >> PAGE_SHIFT);
 257         offset &= ~PAGE_MASK;
 258
 259         while (len > 0) {
 260                 unsigned long bytes;
 261
 262                 bytes = PAGE_SIZE - offset;
 263                 if (bytes > len)
 264                         bytes = len;
 265                 pb[j].pfn = page_to_pfn(page);
 266                 pb[j].offset = offset;
 267                 pb[j].len = bytes;
 268
 269                 offset += bytes;
 270                 len -= bytes;
 271
 272                 if (offset == PAGE_SIZE && len) {
 273                         page++;
 274                         offset = 0;
 275                         j++;
 276                 }
 277         }
 278
 279         return j + 1;
 280 }
 281
 282 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
 283                            struct hv_netvsc_packet *packet,
 284                            struct hv_page_buffer **page_buf)
 285 {
 286         struct hv_page_buffer *pb = *page_buf;
 287         u32 slots_used = 0;
 288         char *data = skb->data;
 289         int frags = skb_shinfo(skb)->nr_frags;
 290         int i;
 291
 292         /* The packet is laid out thus:
 293          * 1. hdr: RNDIS header and PPI
 294          * 2. skb linear data
 295          * 3. skb fragment data
 296          */
 297         if (hdr != NULL)
 298                 slots_used += fill_pg_buf(virt_to_page(hdr),
 299                                         offset_in_page(hdr),
 300                                         len, &pb[slots_used]);
 301
 302         packet->rmsg_size = len;
 303         packet->rmsg_pgcnt = slots_used;
 304
 305         slots_used += fill_pg_buf(virt_to_page(data),
 306                                 offset_in_page(data),
 307                                 skb_headlen(skb), &pb[slots_used]);
 308
 309         for (i = 0; i < frags; i++) {
 310                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 311
 312                 slots_used += fill_pg_buf(skb_frag_page(frag),
 313                                         frag->page_offset,
 314                                         skb_frag_size(frag), &pb[slots_used]);
 315         }
 316         return slots_used;
 317 }
 318
 319 static int count_skb_frag_slots(struct sk_buff *skb)
 320 {
 321         int i, frags = skb_shinfo(skb)->nr_frags;
 322         int pages = 0;
 323
 324         for (i = 0; i < frags; i++) {
 325                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 326                 unsigned long size = skb_frag_size(frag);
 327                 unsigned long offset = frag->page_offset;
 328
 329                 /* Skip unused frames from start of page */
 330                 offset &= ~PAGE_MASK;
 331                 pages += PFN_UP(offset + size);
 332         }
 333         return pages;
 334 }
 335
 336 static int netvsc_get_slots(struct sk_buff *skb)
 337 {
 338         char *data = skb->data;
 339         unsigned int offset = offset_in_page(data);
 340         unsigned int len = skb_headlen(skb);
 341         int slots;
 342         int frag_slots;
 343
 344         slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
 345         frag_slots = count_skb_frag_slots(skb);
 346         return slots + frag_slots;
 347 }
 348
 349 static u32 net_checksum_info(struct sk_buff *skb)
 350 {
 351         if (skb->protocol == htons(ETH_P_IP)) {
 352                 struct iphdr *ip = ip_hdr(skb);
 353
 354                 if (ip->protocol == IPPROTO_TCP)
 355                         return TRANSPORT_INFO_IPV4_TCP;
 356                 else if (ip->protocol == IPPROTO_UDP)
 357                         return TRANSPORT_INFO_IPV4_UDP;
 358         } else {
 359                 struct ipv6hdr *ip6 = ipv6_hdr(skb);
 360
 361                 if (ip6->nexthdr == IPPROTO_TCP)
 362                         return TRANSPORT_INFO_IPV6_TCP;
 363                 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
 364                         return TRANSPORT_INFO_IPV6_UDP;
 365         }
 366
 367         return TRANSPORT_INFO_NOT_IP;
 368 }
 369
 370 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
 371 {
 372         struct net_device_context *net_device_ctx = netdev_priv(net);
 373         struct hv_netvsc_packet *packet = NULL;
 374         int ret;
 375         unsigned int num_data_pgs;
 376         struct rndis_message *rndis_msg;
 377         struct rndis_packet *rndis_pkt;
 378         u32 rndis_msg_size;
 379         struct rndis_per_packet_info *ppi;
 380         u32 hash;
 381         struct hv_page_buffer page_buf[MAX_PAGE_BUFFER_COUNT];
 382         struct hv_page_buffer *pb = page_buf;
 383
 384         /* We will atmost need two pages to describe the rndis
 385          * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
 386          * of pages in a single packet. If skb is scattered around
 387          * more pages we try linearizing it.
 388          */
 389
 390         num_data_pgs = netvsc_get_slots(skb) + 2;
 391
 392         if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
 393                 ++net_device_ctx->eth_stats.tx_scattered;
 394
 395                 if (skb_linearize(skb))
 396                         goto no_memory;
 397
 398                 num_data_pgs = netvsc_get_slots(skb) + 2;
 399                 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
 400                         ++net_device_ctx->eth_stats.tx_too_big;
 401                         goto drop;
 402                 }
 403         }
 404
 405         /*
 406          * Place the rndis header in the skb head room and
 407          * the skb->cb will be used for hv_netvsc_packet
 408          * structure.
 409          */
 410         ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
 411         if (ret)
 412                 goto no_memory;
 413
 414         /* Use the skb control buffer for building up the packet */
 415         BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
 416                         FIELD_SIZEOF(struct sk_buff, cb));
 417         packet = (struct hv_netvsc_packet *)skb->cb;
 418
 419         packet->q_idx = skb_get_queue_mapping(skb);
 420
 421         packet->total_data_buflen = skb->len;
 422         packet->total_bytes = skb->len;
 423         packet->total_packets = 1;
 424
 425         rndis_msg = (struct rndis_message *)skb->head;
 426
 427         memset(rndis_msg, 0, RNDIS_AND_PPI_SIZE);
 428
 429         /* Add the rndis header */
 430         rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
 431         rndis_msg->msg_len = packet->total_data_buflen;
 432         rndis_pkt = &rndis_msg->msg.pkt;
 433         rndis_pkt->data_offset = sizeof(struct rndis_packet);
 434         rndis_pkt->data_len = packet->total_data_buflen;
 435         rndis_pkt->per_pkt_info_offset = sizeof(struct rndis_packet);
 436
 437         rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
 438
 439         hash = skb_get_hash_raw(skb);
 440         if (hash != 0 && net->real_num_tx_queues > 1) {
 441                 rndis_msg_size += NDIS_HASH_PPI_SIZE;
 442                 ppi = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
 443                                     NBL_HASH_VALUE);
 444                 *(u32 *)((void *)ppi + ppi->ppi_offset) = hash;
 445         }
 446
 447         if (skb_vlan_tag_present(skb)) {
 448                 struct ndis_pkt_8021q_info *vlan;
 449
 450                 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
 451                 ppi = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
 452                                         IEEE_8021Q_INFO);
 453                 vlan = (struct ndis_pkt_8021q_info *)((void *)ppi +
 454                                                 ppi->ppi_offset);
 455                 vlan->vlanid = skb->vlan_tci & VLAN_VID_MASK;
 456                 vlan->pri = (skb->vlan_tci & VLAN_PRIO_MASK) >>
 457                                 VLAN_PRIO_SHIFT;
 458         }
 459
 460         if (skb_is_gso(skb)) {
 461                 struct ndis_tcp_lso_info *lso_info;
 462
 463                 rndis_msg_size += NDIS_LSO_PPI_SIZE;
 464                 ppi = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
 465                                     TCP_LARGESEND_PKTINFO);
 466
 467                 lso_info = (struct ndis_tcp_lso_info *)((void *)ppi +
 468                                                         ppi->ppi_offset);
 469
 470                 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
 471                 if (skb->protocol == htons(ETH_P_IP)) {
 472                         lso_info->lso_v2_transmit.ip_version =
 473                                 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
 474                         ip_hdr(skb)->tot_len = 0;
 475                         ip_hdr(skb)->check = 0;
 476                         tcp_hdr(skb)->check =
 477                                 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
 478                                                    ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 479                 } else {
 480                         lso_info->lso_v2_transmit.ip_version =
 481                                 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
 482                         ipv6_hdr(skb)->payload_len = 0;
 483                         tcp_hdr(skb)->check =
 484                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 485                                                  &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 486                 }
 487                 lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb);
 488                 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
 489         } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
 490                 if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) {
 491                         struct ndis_tcp_ip_checksum_info *csum_info;
 492
 493                         rndis_msg_size += NDIS_CSUM_PPI_SIZE;
 494                         ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
 495                                             TCPIP_CHKSUM_PKTINFO);
 496
 497                         csum_info = (struct ndis_tcp_ip_checksum_info *)((void *)ppi +
 498                                                                          ppi->ppi_offset);
 499
 500                         csum_info->transmit.tcp_header_offset = skb_transport_offset(skb);
 501
 502                         if (skb->protocol == htons(ETH_P_IP)) {
 503                                 csum_info->transmit.is_ipv4 = 1;
 504
 505                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
 506                                         csum_info->transmit.tcp_checksum = 1;
 507                                 else
 508                                         csum_info->transmit.udp_checksum = 1;
 509                         } else {
 510                                 csum_info->transmit.is_ipv6 = 1;
 511
 512                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
 513                                         csum_info->transmit.tcp_checksum = 1;
 514                                 else
 515                                         csum_info->transmit.udp_checksum = 1;
 516                         }
 517                 } else {
 518                         /* Can't do offload of this type of checksum */
 519                         if (skb_checksum_help(skb))
 520                                 goto drop;
 521                 }
 522         }
 523
 524         /* Start filling in the page buffers with the rndis hdr */
 525         rndis_msg->msg_len += rndis_msg_size;
 526         packet->total_data_buflen = rndis_msg->msg_len;
 527         packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
 528                                                skb, packet, &pb);
 529
 530         /* timestamp packet in software */
 531         skb_tx_timestamp(skb);
 532         ret = netvsc_send(net_device_ctx->device_ctx, packet,
 533                           rndis_msg, &pb, skb);
 534         if (likely(ret == 0))
 535                 return NETDEV_TX_OK;
 536
 537         if (ret == -EAGAIN) {
 538                 ++net_device_ctx->eth_stats.tx_busy;
 539                 return NETDEV_TX_BUSY;
 540         }
 541
 542         if (ret == -ENOSPC)
 543                 ++net_device_ctx->eth_stats.tx_no_space;
 544
 545 drop:
 546         dev_kfree_skb_any(skb);
 547         net->stats.tx_dropped++;
 548
 549         return NETDEV_TX_OK;
 550
 551 no_memory:
 552         ++net_device_ctx->eth_stats.tx_no_memory;
 553         goto drop;
 554 }
 555 /*
 556  * netvsc_linkstatus_callback - Link up/down notification
 557  */
 558 void netvsc_linkstatus_callback(struct hv_device *device_obj,
 559                                 struct rndis_message *resp)
 560 {
 561         struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
 562         struct net_device *net;
 563         struct net_device_context *ndev_ctx;
 564         struct netvsc_reconfig *event;
 565         unsigned long flags;
 566
 567         net = hv_get_drvdata(device_obj);
 568
 569         if (!net)
 570                 return;
 571
 572         ndev_ctx = netdev_priv(net);
 573
 574         /* Update the physical link speed when changing to another vSwitch */
 575         if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
 576                 u32 speed;
 577
 578                 speed = *(u32 *)((void *)indicate + indicate->
 579                                  status_buf_offset) / 10000;
 580                 ndev_ctx->speed = speed;
 581                 return;
 582         }
 583
 584         /* Handle these link change statuses below */
 585         if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
 586             indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
 587             indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
 588                 return;
 589
 590         if (net->reg_state != NETREG_REGISTERED)
 591                 return;
 592
 593         event = kzalloc(sizeof(*event), GFP_ATOMIC);
 594         if (!event)
 595                 return;
 596         event->event = indicate->status;
 597
 598         spin_lock_irqsave(&ndev_ctx->lock, flags);
 599         list_add_tail(&event->list, &ndev_ctx->reconfig_events);
 600         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
 601
 602         schedule_delayed_work(&ndev_ctx->dwork, 0);
 603 }
 604
 605 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
 606                                              struct napi_struct *napi,
 607                                              const struct ndis_tcp_ip_checksum_info *csum_info,
 608                                              const struct ndis_pkt_8021q_info *vlan,
 609                                              void *data, u32 buflen)
 610 {
 611         struct sk_buff *skb;
 612
 613         skb = napi_alloc_skb(napi, buflen);
 614         if (!skb)
 615                 return skb;
 616
 617         /*
 618          * Copy to skb. This copy is needed here since the memory pointed by
 619          * hv_netvsc_packet cannot be deallocated
 620          */
 621         memcpy(skb_put(skb, buflen), data, buflen);
 622
 623         skb->protocol = eth_type_trans(skb, net);
 624
 625         /* skb is already created with CHECKSUM_NONE */
 626         skb_checksum_none_assert(skb);
 627
 628         /*
 629          * In Linux, the IP checksum is always checked.
 630          * Do L4 checksum offload if enabled and present.
 631          */
 632         if (csum_info && (net->features & NETIF_F_RXCSUM)) {
 633                 if (csum_info->receive.tcp_checksum_succeeded ||
 634                     csum_info->receive.udp_checksum_succeeded)
 635                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 636         }
 637
 638         if (vlan) {
 639                 u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT);
 640
 641                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 642                                        vlan_tci);
 643         }
 644
 645         return skb;
 646 }
 647
 648 /*
 649  * netvsc_recv_callback -  Callback when we receive a packet from the
 650  * "wire" on the specified device.
 651  */
 652 int netvsc_recv_callback(struct net_device *net,
 653                          struct vmbus_channel *channel,
 654                          void  *data, u32 len,
 655                          const struct ndis_tcp_ip_checksum_info *csum_info,
 656                          const struct ndis_pkt_8021q_info *vlan)
 657 {
 658         struct net_device_context *net_device_ctx = netdev_priv(net);
 659         struct netvsc_device *net_device;
 660         u16 q_idx = channel->offermsg.offer.sub_channel_index;
 661         struct netvsc_channel *nvchan;
 662         struct net_device *vf_netdev;
 663         struct sk_buff *skb;
 664         struct netvsc_stats *rx_stats;
 665
 666         if (net->reg_state != NETREG_REGISTERED)
 667                 return NVSP_STAT_FAIL;
 668
 669         /*
 670          * If necessary, inject this packet into the VF interface.
 671          * On Hyper-V, multicast and brodcast packets are only delivered
 672          * to the synthetic interface (after subjecting these to
 673          * policy filters on the host). Deliver these via the VF
 674          * interface in the guest.
 675          */
 676         rcu_read_lock();
 677         net_device = rcu_dereference(net_device_ctx->nvdev);
 678         if (unlikely(!net_device))
 679                 goto drop;
 680
 681         nvchan = &net_device->chan_table[q_idx];
 682         vf_netdev = rcu_dereference(net_device_ctx->vf_netdev);
 683         if (vf_netdev && (vf_netdev->flags & IFF_UP))
 684                 net = vf_netdev;
 685
 686         /* Allocate a skb - TODO direct I/O to pages? */
 687         skb = netvsc_alloc_recv_skb(net, &nvchan->napi,
 688                                     csum_info, vlan, data, len);
 689         if (unlikely(!skb)) {
 690 drop:
 691                 ++net->stats.rx_dropped;
 692                 rcu_read_unlock();
 693                 return NVSP_STAT_FAIL;
 694         }
 695
 696         if (net != vf_netdev)
 697                 skb_record_rx_queue(skb, q_idx);
 698
 699         /*
 700          * Even if injecting the packet, record the statistics
 701          * on the synthetic device because modifying the VF device
 702          * statistics will not work correctly.
 703          */
 704         rx_stats = &nvchan->rx_stats;
 705         u64_stats_update_begin(&rx_stats->syncp);
 706         rx_stats->packets++;
 707         rx_stats->bytes += len;
 708
 709         if (skb->pkt_type == PACKET_BROADCAST)
 710                 ++rx_stats->broadcast;
 711         else if (skb->pkt_type == PACKET_MULTICAST)
 712                 ++rx_stats->multicast;
 713         u64_stats_update_end(&rx_stats->syncp);
 714
 715         napi_gro_receive(&nvchan->napi, skb);
 716         rcu_read_unlock();
 717
 718         return 0;
 719 }
 720
 721 static void netvsc_get_drvinfo(struct net_device *net,
 722                                struct ethtool_drvinfo *info)
 723 {
 724         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
 725         strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
 726 }
 727
 728 static void netvsc_get_channels(struct net_device *net,
 729                                 struct ethtool_channels *channel)
 730 {
 731         struct net_device_context *net_device_ctx = netdev_priv(net);
 732         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
 733
 734         if (nvdev) {
 735                 channel->max_combined   = nvdev->max_chn;
 736                 channel->combined_count = nvdev->num_chn;
 737         }
 738 }
 739
 740 static int netvsc_set_queues(struct net_device *net, struct hv_device *dev,
 741                              u32 num_chn)
 742 {
 743         struct netvsc_device_info device_info;
 744         int ret;
 745
 746         memset(&device_info, 0, sizeof(device_info));
 747         device_info.num_chn = num_chn;
 748         device_info.ring_size = ring_size;
 749         device_info.max_num_vrss_chns = num_chn;
 750
 751         ret = rndis_filter_device_add(dev, &device_info);
 752         if (ret)
 753                 return ret;
 754
 755         ret = netif_set_real_num_tx_queues(net, num_chn);
 756         if (ret)
 757                 return ret;
 758
 759         ret = netif_set_real_num_rx_queues(net, num_chn);
 760
 761         return ret;
 762 }
 763
 764 static int netvsc_set_channels(struct net_device *net,
 765                                struct ethtool_channels *channels)
 766 {
 767         struct net_device_context *net_device_ctx = netdev_priv(net);
 768         struct hv_device *dev = net_device_ctx->device_ctx;
 769         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
 770         unsigned int count = channels->combined_count;
 771         bool was_running;
 772         int ret;
 773
 774         /* We do not support separate count for rx, tx, or other */
 775         if (count == 0 ||
 776             channels->rx_count || channels->tx_count || channels->other_count)
 777                 return -EINVAL;
 778
 779         if (count > net->num_tx_queues || count > VRSS_CHANNEL_MAX)
 780                 return -EINVAL;
 781
 782         if (!nvdev || nvdev->destroy)
 783                 return -ENODEV;
 784
 785         if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
 786                 return -EINVAL;
 787
 788         if (count > nvdev->max_chn)
 789                 return -EINVAL;
 790
 791         was_running = netif_running(net);
 792         if (was_running) {
 793                 ret = netvsc_close(net);
 794                 if (ret)
 795                         return ret;
 796         }
 797
 798         rndis_filter_device_remove(dev, nvdev);
 799
 800         ret = netvsc_set_queues(net, dev, count);
 801         if (ret == 0)
 802                 nvdev->num_chn = count;
 803         else
 804                 netvsc_set_queues(net, dev, nvdev->num_chn);
 805
 806         if (was_running)
 807                 ret = netvsc_open(net);
 808
 809         /* We may have missed link change notifications */
 810         schedule_delayed_work(&net_device_ctx->dwork, 0);
 811
 812         return ret;
 813 }
 814
 815 static bool
 816 netvsc_validate_ethtool_ss_cmd(const struct ethtool_link_ksettings *cmd)
 817 {
 818         struct ethtool_link_ksettings diff1 = *cmd;
 819         struct ethtool_link_ksettings diff2 = {};
 820
 821         diff1.base.speed = 0;
 822         diff1.base.duplex = 0;
 823         /* advertising and cmd are usually set */
 824         ethtool_link_ksettings_zero_link_mode(&diff1, advertising);
 825         diff1.base.cmd = 0;
 826         /* We set port to PORT_OTHER */
 827         diff2.base.port = PORT_OTHER;
 828
 829         return !memcmp(&diff1, &diff2, sizeof(diff1));
 830 }
 831
 832 static void netvsc_init_settings(struct net_device *dev)
 833 {
 834         struct net_device_context *ndc = netdev_priv(dev);
 835
 836         ndc->speed = SPEED_UNKNOWN;
 837         ndc->duplex = DUPLEX_FULL;
 838 }
 839
 840 static int netvsc_get_link_ksettings(struct net_device *dev,
 841                                      struct ethtool_link_ksettings *cmd)
 842 {
 843         struct net_device_context *ndc = netdev_priv(dev);
 844
 845         cmd->base.speed = ndc->speed;
 846         cmd->base.duplex = ndc->duplex;
 847         cmd->base.port = PORT_OTHER;
 848
 849         return 0;
 850 }
 851
 852 static int netvsc_set_link_ksettings(struct net_device *dev,
 853                                      const struct ethtool_link_ksettings *cmd)
 854 {
 855         struct net_device_context *ndc = netdev_priv(dev);
 856         u32 speed;
 857
 858         speed = cmd->base.speed;
 859         if (!ethtool_validate_speed(speed) ||
 860             !ethtool_validate_duplex(cmd->base.duplex) ||
 861             !netvsc_validate_ethtool_ss_cmd(cmd))
 862                 return -EINVAL;
 863
 864         ndc->speed = speed;
 865         ndc->duplex = cmd->base.duplex;
 866
 867         return 0;
 868 }
 869
 870 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
 871 {
 872         struct net_device_context *ndevctx = netdev_priv(ndev);
 873         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
 874         struct hv_device *hdev = ndevctx->device_ctx;
 875         struct netvsc_device_info device_info;
 876         bool was_running;
 877         int ret = 0;
 878
 879         if (!nvdev || nvdev->destroy)
 880                 return -ENODEV;
 881
 882         was_running = netif_running(ndev);
 883         if (was_running) {
 884                 ret = netvsc_close(ndev);
 885                 if (ret)
 886                         return ret;
 887         }
 888
 889         memset(&device_info, 0, sizeof(device_info));
 890         device_info.ring_size = ring_size;
 891         device_info.num_chn = nvdev->num_chn;
 892         device_info.max_num_vrss_chns = nvdev->num_chn;
 893
 894         rndis_filter_device_remove(hdev, nvdev);
 895
 896         /* 'nvdev' has been freed in rndis_filter_device_remove() ->
 897          * netvsc_device_remove () -> free_netvsc_device().
 898          * We mustn't access it before it's re-created in
 899          * rndis_filter_device_add() -> netvsc_device_add().
 900          */
 901
 902         ndev->mtu = mtu;
 903
 904         rndis_filter_device_add(hdev, &device_info);
 905
 906         if (was_running)
 907                 ret = netvsc_open(ndev);
 908
 909         /* We may have missed link change notifications */
 910         schedule_delayed_work(&ndevctx->dwork, 0);
 911
 912         return ret;
 913 }
 914
 915 static void netvsc_get_stats64(struct net_device *net,
 916                                struct rtnl_link_stats64 *t)
 917 {
 918         struct net_device_context *ndev_ctx = netdev_priv(net);
 919         struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
 920         int i;
 921
 922         if (!nvdev)
 923                 return;
 924
 925         for (i = 0; i < nvdev->num_chn; i++) {
 926                 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
 927                 const struct netvsc_stats *stats;
 928                 u64 packets, bytes, multicast;
 929                 unsigned int start;
 930
 931                 stats = &nvchan->tx_stats;
 932                 do {
 933                         start = u64_stats_fetch_begin_irq(&stats->syncp);
 934                         packets = stats->packets;
 935                         bytes = stats->bytes;
 936                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
 937
 938                 t->tx_bytes     += bytes;
 939                 t->tx_packets   += packets;
 940
 941                 stats = &nvchan->rx_stats;
 942                 do {
 943                         start = u64_stats_fetch_begin_irq(&stats->syncp);
 944                         packets = stats->packets;
 945                         bytes = stats->bytes;
 946                         multicast = stats->multicast + stats->broadcast;
 947                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
 948
 949                 t->rx_bytes     += bytes;
 950                 t->rx_packets   += packets;
 951                 t->multicast    += multicast;
 952         }
 953
 954         t->tx_dropped   = net->stats.tx_dropped;
 955         t->tx_errors    = net->stats.tx_errors;
 956
 957         t->rx_dropped   = net->stats.rx_dropped;
 958         t->rx_errors    = net->stats.rx_errors;
 959 }
 960
 961 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
 962 {
 963         struct sockaddr *addr = p;
 964         char save_adr[ETH_ALEN];
 965         unsigned char save_aatype;
 966         int err;
 967
 968         memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
 969         save_aatype = ndev->addr_assign_type;
 970
 971         err = eth_mac_addr(ndev, p);
 972         if (err != 0)
 973                 return err;
 974
 975         err = rndis_filter_set_device_mac(ndev, addr->sa_data);
 976         if (err != 0) {
 977                 /* roll back to saved MAC */
 978                 memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
 979                 ndev->addr_assign_type = save_aatype;
 980         }
 981
 982         return err;
 983 }
 984
 985 static const struct {
 986         char name[ETH_GSTRING_LEN];
 987         u16 offset;
 988 } netvsc_stats[] = {
 989         { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
 990         { "tx_no_memory",  offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
 991         { "tx_no_space",  offsetof(struct netvsc_ethtool_stats, tx_no_space) },
 992         { "tx_too_big",   offsetof(struct netvsc_ethtool_stats, tx_too_big) },
 993         { "tx_busy",      offsetof(struct netvsc_ethtool_stats, tx_busy) },
 994 };
 995
 996 #define NETVSC_GLOBAL_STATS_LEN ARRAY_SIZE(netvsc_stats)
 997
 998 /* 4 statistics per queue (rx/tx packets/bytes) */
 999 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 4)
1000
1001 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
1002 {
1003         struct net_device_context *ndc = netdev_priv(dev);
1004         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1005
1006         if (!nvdev)
1007                 return -ENODEV;
1008
1009         switch (string_set) {
1010         case ETH_SS_STATS:
1011                 return NETVSC_GLOBAL_STATS_LEN + NETVSC_QUEUE_STATS_LEN(nvdev);
1012         default:
1013                 return -EINVAL;
1014         }
1015 }
1016
1017 static void netvsc_get_ethtool_stats(struct net_device *dev,
1018                                      struct ethtool_stats *stats, u64 *data)
1019 {
1020         struct net_device_context *ndc = netdev_priv(dev);
1021         struct netvsc_device *nvdev = rcu_dereference(ndc->nvdev);
1022         const void *nds = &ndc->eth_stats;
1023         const struct netvsc_stats *qstats;
1024         unsigned int start;
1025         u64 packets, bytes;
1026         int i, j;
1027
1028         if (!nvdev)
1029                 return;
1030
1031         for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
1032                 data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1033
1034         for (j = 0; j < nvdev->num_chn; j++) {
1035                 qstats = &nvdev->chan_table[j].tx_stats;
1036
1037                 do {
1038                         start = u64_stats_fetch_begin_irq(&qstats->syncp);
1039                         packets = qstats->packets;
1040                         bytes = qstats->bytes;
1041                 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1042                 data[i++] = packets;
1043                 data[i++] = bytes;
1044
1045                 qstats = &nvdev->chan_table[j].rx_stats;
1046                 do {
1047                         start = u64_stats_fetch_begin_irq(&qstats->syncp);
1048                         packets = qstats->packets;
1049                         bytes = qstats->bytes;
1050                 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1051                 data[i++] = packets;
1052                 data[i++] = bytes;
1053         }
1054 }
1055
1056 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1057 {
1058         struct net_device_context *ndc = netdev_priv(dev);
1059         struct netvsc_device *nvdev = rcu_dereference(ndc->nvdev);
1060         u8 *p = data;
1061         int i;
1062
1063         if (!nvdev)
1064                 return;
1065
1066         switch (stringset) {
1067         case ETH_SS_STATS:
1068                 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++)
1069                         memcpy(p + i * ETH_GSTRING_LEN,
1070                                netvsc_stats[i].name, ETH_GSTRING_LEN);
1071
1072                 p += i * ETH_GSTRING_LEN;
1073                 for (i = 0; i < nvdev->num_chn; i++) {
1074                         sprintf(p, "tx_queue_%u_packets", i);
1075                         p += ETH_GSTRING_LEN;
1076                         sprintf(p, "tx_queue_%u_bytes", i);
1077                         p += ETH_GSTRING_LEN;
1078                         sprintf(p, "rx_queue_%u_packets", i);
1079                         p += ETH_GSTRING_LEN;
1080                         sprintf(p, "rx_queue_%u_bytes", i);
1081                         p += ETH_GSTRING_LEN;
1082                 }
1083
1084                 break;
1085         }
1086 }
1087
1088 static int
1089 netvsc_get_rss_hash_opts(struct netvsc_device *nvdev,
1090                          struct ethtool_rxnfc *info)
1091 {
1092         info->data = RXH_IP_SRC | RXH_IP_DST;
1093
1094         switch (info->flow_type) {
1095         case TCP_V4_FLOW:
1096         case TCP_V6_FLOW:
1097                 info->data |= RXH_L4_B_0_1 | RXH_L4_B_2_3;
1098                 /* fallthrough */
1099         case UDP_V4_FLOW:
1100         case UDP_V6_FLOW:
1101         case IPV4_FLOW:
1102         case IPV6_FLOW:
1103                 break;
1104         default:
1105                 info->data = 0;
1106                 break;
1107         }
1108
1109         return 0;
1110 }
1111
1112 static int
1113 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1114                  u32 *rules)
1115 {
1116         struct net_device_context *ndc = netdev_priv(dev);
1117         struct netvsc_device *nvdev = rcu_dereference(ndc->nvdev);
1118
1119         if (!nvdev)
1120                 return -ENODEV;
1121
1122         switch (info->cmd) {
1123         case ETHTOOL_GRXRINGS:
1124                 info->data = nvdev->num_chn;
1125                 return 0;
1126
1127         case ETHTOOL_GRXFH:
1128                 return netvsc_get_rss_hash_opts(nvdev, info);
1129         }
1130         return -EOPNOTSUPP;
1131 }
1132
1133 #ifdef CONFIG_NET_POLL_CONTROLLER
1134 static void netvsc_poll_controller(struct net_device *dev)
1135 {
1136         struct net_device_context *ndc = netdev_priv(dev);
1137         struct netvsc_device *ndev;
1138         int i;
1139
1140         rcu_read_lock();
1141         ndev = rcu_dereference(ndc->nvdev);
1142         if (ndev) {
1143                 for (i = 0; i < ndev->num_chn; i++) {
1144                         struct netvsc_channel *nvchan = &ndev->chan_table[i];
1145
1146                         napi_schedule(&nvchan->napi);
1147                 }
1148         }
1149         rcu_read_unlock();
1150 }
1151 #endif
1152
1153 static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1154 {
1155         return NETVSC_HASH_KEYLEN;
1156 }
1157
1158 static u32 netvsc_rss_indir_size(struct net_device *dev)
1159 {
1160         return ITAB_NUM;
1161 }
1162
1163 static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key,
1164                            u8 *hfunc)
1165 {
1166         struct net_device_context *ndc = netdev_priv(dev);
1167         struct netvsc_device *ndev = rcu_dereference(ndc->nvdev);
1168         struct rndis_device *rndis_dev;
1169         int i;
1170
1171         if (!ndev)
1172                 return -ENODEV;
1173
1174         if (hfunc)
1175                 *hfunc = ETH_RSS_HASH_TOP;      /* Toeplitz */
1176
1177         rndis_dev = ndev->extension;
1178         if (indir) {
1179                 for (i = 0; i < ITAB_NUM; i++)
1180                         indir[i] = rndis_dev->ind_table[i];
1181         }
1182
1183         if (key)
1184                 memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1185
1186         return 0;
1187 }
1188
1189 static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
1190                            const u8 *key, const u8 hfunc)
1191 {
1192         struct net_device_context *ndc = netdev_priv(dev);
1193         struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1194         struct rndis_device *rndis_dev;
1195         int i;
1196
1197         if (!ndev)
1198                 return -ENODEV;
1199
1200         if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
1201                 return -EOPNOTSUPP;
1202
1203         rndis_dev = ndev->extension;
1204         if (indir) {
1205                 for (i = 0; i < ITAB_NUM; i++)
1206                         if (indir[i] >= VRSS_CHANNEL_MAX)
1207                                 return -EINVAL;
1208
1209                 for (i = 0; i < ITAB_NUM; i++)
1210                         rndis_dev->ind_table[i] = indir[i];
1211         }
1212
1213         if (!key) {
1214                 if (!indir)
1215                         return 0;
1216
1217                 key = rndis_dev->rss_key;
1218         }
1219
1220         return rndis_filter_set_rss_param(rndis_dev, key, ndev->num_chn);
1221 }
1222
1223 static const struct ethtool_ops ethtool_ops = {
1224         .get_drvinfo    = netvsc_get_drvinfo,
1225         .get_link       = ethtool_op_get_link,
1226         .get_ethtool_stats = netvsc_get_ethtool_stats,
1227         .get_sset_count = netvsc_get_sset_count,
1228         .get_strings    = netvsc_get_strings,
1229         .get_channels   = netvsc_get_channels,
1230         .set_channels   = netvsc_set_channels,
1231         .get_ts_info    = ethtool_op_get_ts_info,
1232         .get_rxnfc      = netvsc_get_rxnfc,
1233         .get_rxfh_key_size = netvsc_get_rxfh_key_size,
1234         .get_rxfh_indir_size = netvsc_rss_indir_size,
1235         .get_rxfh       = netvsc_get_rxfh,
1236         .set_rxfh       = netvsc_set_rxfh,
1237         .get_link_ksettings = netvsc_get_link_ksettings,
1238         .set_link_ksettings = netvsc_set_link_ksettings,
1239 };
1240
1241 static const struct net_device_ops device_ops = {
1242         .ndo_open =                     netvsc_open,
1243         .ndo_stop =                     netvsc_close,
1244         .ndo_start_xmit =               netvsc_start_xmit,
1245         .ndo_set_rx_mode =              netvsc_set_multicast_list,
1246         .ndo_change_mtu =               netvsc_change_mtu,
1247         .ndo_validate_addr =            eth_validate_addr,
1248         .ndo_set_mac_address =          netvsc_set_mac_addr,
1249         .ndo_select_queue =             netvsc_select_queue,
1250         .ndo_get_stats64 =              netvsc_get_stats64,
1251 #ifdef CONFIG_NET_POLL_CONTROLLER
1252         .ndo_poll_controller =          netvsc_poll_controller,
1253 #endif
1254 };
1255
1256 /*
1257  * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
1258  * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
1259  * present send GARP packet to network peers with netif_notify_peers().
1260  */
1261 static void netvsc_link_change(struct work_struct *w)
1262 {
1263         struct net_device_context *ndev_ctx =
1264                 container_of(w, struct net_device_context, dwork.work);
1265         struct hv_device *device_obj = ndev_ctx->device_ctx;
1266         struct net_device *net = hv_get_drvdata(device_obj);
1267         struct netvsc_device *net_device;
1268         struct rndis_device *rdev;
1269         struct netvsc_reconfig *event = NULL;
1270         bool notify = false, reschedule = false;
1271         unsigned long flags, next_reconfig, delay;
1272
1273         rtnl_lock();
1274         net_device = rtnl_dereference(ndev_ctx->nvdev);
1275         if (!net_device)
1276                 goto out_unlock;
1277
1278         rdev = net_device->extension;
1279
1280         next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
1281         if (time_is_after_jiffies(next_reconfig)) {
1282                 /* link_watch only sends one notification with current state
1283                  * per second, avoid doing reconfig more frequently. Handle
1284                  * wrap around.
1285                  */
1286                 delay = next_reconfig - jiffies;
1287                 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
1288                 schedule_delayed_work(&ndev_ctx->dwork, delay);
1289                 goto out_unlock;
1290         }
1291         ndev_ctx->last_reconfig = jiffies;
1292
1293         spin_lock_irqsave(&ndev_ctx->lock, flags);
1294         if (!list_empty(&ndev_ctx->reconfig_events)) {
1295                 event = list_first_entry(&ndev_ctx->reconfig_events,
1296                                          struct netvsc_reconfig, list);
1297                 list_del(&event->list);
1298                 reschedule = !list_empty(&ndev_ctx->reconfig_events);
1299         }
1300         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1301
1302         if (!event)
1303                 goto out_unlock;
1304
1305         switch (event->event) {
1306                 /* Only the following events are possible due to the check in
1307                  * netvsc_linkstatus_callback()
1308                  */
1309         case RNDIS_STATUS_MEDIA_CONNECT:
1310                 if (rdev->link_state) {
1311                         rdev->link_state = false;
1312                         netif_carrier_on(net);
1313                         netif_tx_wake_all_queues(net);
1314                 } else {
1315                         notify = true;
1316                 }
1317                 kfree(event);
1318                 break;
1319         case RNDIS_STATUS_MEDIA_DISCONNECT:
1320                 if (!rdev->link_state) {
1321                         rdev->link_state = true;
1322                         netif_carrier_off(net);
1323                         netif_tx_stop_all_queues(net);
1324                 }
1325                 kfree(event);
1326                 break;
1327         case RNDIS_STATUS_NETWORK_CHANGE:
1328                 /* Only makes sense if carrier is present */
1329                 if (!rdev->link_state) {
1330                         rdev->link_state = true;
1331                         netif_carrier_off(net);
1332                         netif_tx_stop_all_queues(net);
1333                         event->event = RNDIS_STATUS_MEDIA_CONNECT;
1334                         spin_lock_irqsave(&ndev_ctx->lock, flags);
1335                         list_add(&event->list, &ndev_ctx->reconfig_events);
1336                         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1337                         reschedule = true;
1338                 }
1339                 break;
1340         }
1341
1342         rtnl_unlock();
1343
1344         if (notify)
1345                 netdev_notify_peers(net);
1346
1347         /* link_watch only sends one notification with current state per
1348          * second, handle next reconfig event in 2 seconds.
1349          */
1350         if (reschedule)
1351                 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1352
1353         return;
1354
1355 out_unlock:
1356         rtnl_unlock();
1357 }
1358
1359 static struct net_device *get_netvsc_bymac(const u8 *mac)
1360 {
1361         struct net_device *dev;
1362
1363         ASSERT_RTNL();
1364
1365         for_each_netdev(&init_net, dev) {
1366                 if (dev->netdev_ops != &device_ops)
1367                         continue;       /* not a netvsc device */
1368
1369                 if (ether_addr_equal(mac, dev->perm_addr))
1370                         return dev;
1371         }
1372
1373         return NULL;
1374 }
1375
1376 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
1377 {
1378         struct net_device *dev;
1379
1380         ASSERT_RTNL();
1381
1382         for_each_netdev(&init_net, dev) {
1383                 struct net_device_context *net_device_ctx;
1384
1385                 if (dev->netdev_ops != &device_ops)
1386                         continue;       /* not a netvsc device */
1387
1388                 net_device_ctx = netdev_priv(dev);
1389                 if (net_device_ctx->nvdev == NULL)
1390                         continue;       /* device is removed */
1391
1392                 if (rtnl_dereference(net_device_ctx->vf_netdev) == vf_netdev)
1393                         return dev;     /* a match */
1394         }
1395
1396         return NULL;
1397 }
1398
1399 static int netvsc_register_vf(struct net_device *vf_netdev)
1400 {
1401         struct net_device *ndev;
1402         struct net_device_context *net_device_ctx;
1403         struct netvsc_device *netvsc_dev;
1404
1405         if (vf_netdev->addr_len != ETH_ALEN)
1406                 return NOTIFY_DONE;
1407
1408         /*
1409          * We will use the MAC address to locate the synthetic interface to
1410          * associate with the VF interface. If we don't find a matching
1411          * synthetic interface, move on.
1412          */
1413         ndev = get_netvsc_bymac(vf_netdev->perm_addr);
1414         if (!ndev)
1415                 return NOTIFY_DONE;
1416
1417         net_device_ctx = netdev_priv(ndev);
1418         netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
1419         if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
1420                 return NOTIFY_DONE;
1421
1422         netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
1423         /*
1424          * Take a reference on the module.
1425          */
1426         try_module_get(THIS_MODULE);
1427
1428         dev_hold(vf_netdev);
1429         rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
1430         return NOTIFY_OK;
1431 }
1432
1433 static int netvsc_vf_up(struct net_device *vf_netdev)
1434 {
1435         struct net_device *ndev;
1436         struct netvsc_device *netvsc_dev;
1437         struct net_device_context *net_device_ctx;
1438
1439         ndev = get_netvsc_byref(vf_netdev);
1440         if (!ndev)
1441                 return NOTIFY_DONE;
1442
1443         net_device_ctx = netdev_priv(ndev);
1444         netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
1445
1446         netdev_info(ndev, "VF up: %s\n", vf_netdev->name);
1447
1448         /*
1449          * Open the device before switching data path.
1450          */
1451         rndis_filter_open(netvsc_dev);
1452
1453         /*
1454          * notify the host to switch the data path.
1455          */
1456         netvsc_switch_datapath(ndev, true);
1457         netdev_info(ndev, "Data path switched to VF: %s\n", vf_netdev->name);
1458
1459         netif_carrier_off(ndev);
1460
1461         /* Now notify peers through VF device. */
1462         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, vf_netdev);
1463
1464         return NOTIFY_OK;
1465 }
1466
1467 static int netvsc_vf_down(struct net_device *vf_netdev)
1468 {
1469         struct net_device *ndev;
1470         struct netvsc_device *netvsc_dev;
1471         struct net_device_context *net_device_ctx;
1472
1473         ndev = get_netvsc_byref(vf_netdev);
1474         if (!ndev)
1475                 return NOTIFY_DONE;
1476
1477         net_device_ctx = netdev_priv(ndev);
1478         netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
1479
1480         netdev_info(ndev, "VF down: %s\n", vf_netdev->name);
1481         netvsc_switch_datapath(ndev, false);
1482         netdev_info(ndev, "Data path switched from VF: %s\n", vf_netdev->name);
1483         rndis_filter_close(netvsc_dev);
1484         netif_carrier_on(ndev);
1485
1486         /* Now notify peers through netvsc device. */
1487         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, ndev);
1488
1489         return NOTIFY_OK;
1490 }
1491
1492 static int netvsc_unregister_vf(struct net_device *vf_netdev)
1493 {
1494         struct net_device *ndev;
1495         struct net_device_context *net_device_ctx;
1496
1497         ndev = get_netvsc_byref(vf_netdev);
1498         if (!ndev)
1499                 return NOTIFY_DONE;
1500
1501         net_device_ctx = netdev_priv(ndev);
1502
1503         netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
1504
1505         RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
1506         dev_put(vf_netdev);
1507         module_put(THIS_MODULE);
1508         return NOTIFY_OK;
1509 }
1510
1511 static int netvsc_probe(struct hv_device *dev,
1512                         const struct hv_vmbus_device_id *dev_id)
1513 {
1514         struct net_device *net = NULL;
1515         struct net_device_context *net_device_ctx;
1516         struct netvsc_device_info device_info;
1517         struct netvsc_device *nvdev;
1518         int ret;
1519
1520         net = alloc_etherdev_mq(sizeof(struct net_device_context),
1521                                 VRSS_CHANNEL_MAX);
1522         if (!net)
1523                 return -ENOMEM;
1524
1525         netif_carrier_off(net);
1526
1527         netvsc_init_settings(net);
1528
1529         net_device_ctx = netdev_priv(net);
1530         net_device_ctx->device_ctx = dev;
1531         net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
1532         if (netif_msg_probe(net_device_ctx))
1533                 netdev_dbg(net, "netvsc msg_enable: %d\n",
1534                            net_device_ctx->msg_enable);
1535
1536         hv_set_drvdata(dev, net);
1537
1538         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
1539
1540         spin_lock_init(&net_device_ctx->lock);
1541         INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
1542
1543         net->netdev_ops = &device_ops;
1544         net->ethtool_ops = &ethtool_ops;
1545         SET_NETDEV_DEV(net, &dev->device);
1546
1547         /* We always need headroom for rndis header */
1548         net->needed_headroom = RNDIS_AND_PPI_SIZE;
1549
1550         /* Notify the netvsc driver of the new device */
1551         memset(&device_info, 0, sizeof(device_info));
1552         device_info.ring_size = ring_size;
1553         device_info.num_chn = VRSS_CHANNEL_DEFAULT;
1554         ret = rndis_filter_device_add(dev, &device_info);
1555         if (ret != 0) {
1556                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
1557                 free_netdev(net);
1558                 hv_set_drvdata(dev, NULL);
1559                 return ret;
1560         }
1561         memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
1562
1563         /* hw_features computed in rndis_filter_device_add */
1564         net->features = net->hw_features |
1565                 NETIF_F_HIGHDMA | NETIF_F_SG |
1566                 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
1567         net->vlan_features = net->features;
1568
1569         /* RCU not necessary here, device not registered */
1570         nvdev = net_device_ctx->nvdev;
1571         netif_set_real_num_tx_queues(net, nvdev->num_chn);
1572         netif_set_real_num_rx_queues(net, nvdev->num_chn);
1573
1574         /* MTU range: 68 - 1500 or 65521 */
1575         net->min_mtu = NETVSC_MTU_MIN;
1576         if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
1577                 net->max_mtu = NETVSC_MTU - ETH_HLEN;
1578         else
1579                 net->max_mtu = ETH_DATA_LEN;
1580
1581         ret = register_netdev(net);
1582         if (ret != 0) {
1583                 pr_err("Unable to register netdev.\n");
1584                 rndis_filter_device_remove(dev, nvdev);
1585                 free_netdev(net);
1586         }
1587
1588         return ret;
1589 }
1590
1591 static int netvsc_remove(struct hv_device *dev)
1592 {
1593         struct net_device *net;
1594         struct net_device_context *ndev_ctx;
1595
1596         net = hv_get_drvdata(dev);
1597
1598         if (net == NULL) {
1599                 dev_err(&dev->device, "No net device to remove\n");
1600                 return 0;
1601         }
1602
1603         ndev_ctx = netdev_priv(net);
1604
1605         netif_device_detach(net);
1606
1607         cancel_delayed_work_sync(&ndev_ctx->dwork);
1608
1609         /*
1610          * Call to the vsc driver to let it know that the device is being
1611          * removed. Also blocks mtu and channel changes.
1612          */
1613         rtnl_lock();
1614         rndis_filter_device_remove(dev, ndev_ctx->nvdev);
1615         rtnl_unlock();
1616
1617         unregister_netdev(net);
1618
1619         hv_set_drvdata(dev, NULL);
1620
1621         free_netdev(net);
1622         return 0;
1623 }
1624
1625 static const struct hv_vmbus_device_id id_table[] = {
1626         /* Network guid */
1627         { HV_NIC_GUID, },
1628         { },
1629 };
1630
1631 MODULE_DEVICE_TABLE(vmbus, id_table);
1632
1633 /* The one and only one */
1634 static struct  hv_driver netvsc_drv = {
1635         .name = KBUILD_MODNAME,
1636         .id_table = id_table,
1637         .probe = netvsc_probe,
1638         .remove = netvsc_remove,
1639 };
1640
1641 /*
1642  * On Hyper-V, every VF interface is matched with a corresponding
1643  * synthetic interface. The synthetic interface is presented first
1644  * to the guest. When the corresponding VF instance is registered,
1645  * we will take care of switching the data path.
1646  */
1647 static int netvsc_netdev_event(struct notifier_block *this,
1648                                unsigned long event, void *ptr)
1649 {
1650         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
1651
1652         /* Skip our own events */
1653         if (event_dev->netdev_ops == &device_ops)
1654                 return NOTIFY_DONE;
1655
1656         /* Avoid non-Ethernet type devices */
1657         if (event_dev->type != ARPHRD_ETHER)
1658                 return NOTIFY_DONE;
1659
1660         /* Avoid Vlan dev with same MAC registering as VF */
1661         if (is_vlan_dev(event_dev))
1662                 return NOTIFY_DONE;
1663
1664         /* Avoid Bonding master dev with same MAC registering as VF */
1665         if ((event_dev->priv_flags & IFF_BONDING) &&
1666             (event_dev->flags & IFF_MASTER))
1667                 return NOTIFY_DONE;
1668
1669         switch (event) {
1670         case NETDEV_REGISTER:
1671                 return netvsc_register_vf(event_dev);
1672         case NETDEV_UNREGISTER:
1673                 return netvsc_unregister_vf(event_dev);
1674         case NETDEV_UP:
1675                 return netvsc_vf_up(event_dev);
1676         case NETDEV_DOWN:
1677                 return netvsc_vf_down(event_dev);
1678         default:
1679                 return NOTIFY_DONE;
1680         }
1681 }
1682
1683 static struct notifier_block netvsc_netdev_notifier = {
1684         .notifier_call = netvsc_netdev_event,
1685 };
1686
1687 static void __exit netvsc_drv_exit(void)
1688 {
1689         unregister_netdevice_notifier(&netvsc_netdev_notifier);
1690         vmbus_driver_unregister(&netvsc_drv);
1691 }
1692
1693 static int __init netvsc_drv_init(void)
1694 {
1695         int ret;
1696
1697         if (ring_size < RING_SIZE_MIN) {
1698                 ring_size = RING_SIZE_MIN;
1699                 pr_info("Increased ring_size to %d (min allowed)\n",
1700                         ring_size);
1701         }
1702         ret = vmbus_driver_register(&netvsc_drv);
1703
1704         if (ret)
1705                 return ret;
1706
1707         register_netdevice_notifier(&netvsc_netdev_notifier);
1708         return 0;
1709 }
1710
1711 MODULE_LICENSE("GPL");
1712 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
1713
1714 module_init(netvsc_drv_init);
1715 module_exit(netvsc_drv_exit);