drivers/net/hyperv/netvsc_drv.c

   1 /*
   2  * Copyright (c) 2009, Microsoft Corporation.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms and conditions of the GNU General Public License,
   6  * version 2, as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope it will be useful, but WITHOUT
   9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  11  * more details.
  12  *
  13  * You should have received a copy of the GNU General Public License along with
  14  * this program; if not, see <http://www.gnu.org/licenses/>.
  15  *
  16  * Authors:
  17  *   Haiyang Zhang <haiyangz@microsoft.com>
  18  *   Hank Janssen  <hjanssen@microsoft.com>
  19  */
  20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  21
  22 #include <linux/init.h>
  23 #include <linux/atomic.h>
  24 #include <linux/module.h>
  25 #include <linux/highmem.h>
  26 #include <linux/device.h>
  27 #include <linux/io.h>
  28 #include <linux/delay.h>
  29 #include <linux/netdevice.h>
  30 #include <linux/inetdevice.h>
  31 #include <linux/etherdevice.h>
  32 #include <linux/skbuff.h>
  33 #include <linux/if_vlan.h>
  34 #include <linux/in.h>
  35 #include <linux/slab.h>
  36 #include <net/arp.h>
  37 #include <net/route.h>
  38 #include <net/sock.h>
  39 #include <net/pkt_sched.h>
  40
  41 #include "hyperv_net.h"
  42
  43 /* Restrict GSO size to account for NVGRE */
  44 #define NETVSC_GSO_MAX_SIZE     62768
  45
  46 #define RING_SIZE_MIN 64
  47 static int ring_size = 128;
  48 module_param(ring_size, int, S_IRUGO);
  49 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
  50
  51 static int max_num_vrss_chns = 8;
  52
  53 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
  54                                 NETIF_MSG_LINK | NETIF_MSG_IFUP |
  55                                 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
  56                                 NETIF_MSG_TX_ERR;
  57
  58 static int debug = -1;
  59 module_param(debug, int, S_IRUGO);
  60 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  61
  62 static void do_set_multicast(struct work_struct *w)
  63 {
  64         struct net_device_context *ndevctx =
  65                 container_of(w, struct net_device_context, work);
  66         struct netvsc_device *nvdev;
  67         struct rndis_device *rdev;
  68
  69         nvdev = hv_get_drvdata(ndevctx->device_ctx);
  70         if (nvdev == NULL || nvdev->ndev == NULL)
  71                 return;
  72
  73         rdev = nvdev->extension;
  74         if (rdev == NULL)
  75                 return;
  76
  77         if (nvdev->ndev->flags & IFF_PROMISC)
  78                 rndis_filter_set_packet_filter(rdev,
  79                         NDIS_PACKET_TYPE_PROMISCUOUS);
  80         else
  81                 rndis_filter_set_packet_filter(rdev,
  82                         NDIS_PACKET_TYPE_BROADCAST |
  83                         NDIS_PACKET_TYPE_ALL_MULTICAST |
  84                         NDIS_PACKET_TYPE_DIRECTED);
  85 }
  86
  87 static void netvsc_set_multicast_list(struct net_device *net)
  88 {
  89         struct net_device_context *net_device_ctx = netdev_priv(net);
  90
  91         schedule_work(&net_device_ctx->work);
  92 }
  93
  94 static int netvsc_open(struct net_device *net)
  95 {
  96         struct net_device_context *net_device_ctx = netdev_priv(net);
  97         struct hv_device *device_obj = net_device_ctx->device_ctx;
  98         struct netvsc_device *nvdev;
  99         struct rndis_device *rdev;
 100         int ret = 0;
 101
 102         netif_carrier_off(net);
 103
 104         /* Open up the device */
 105         ret = rndis_filter_open(device_obj);
 106         if (ret != 0) {
 107                 netdev_err(net, "unable to open device (ret %d).\n", ret);
 108                 return ret;
 109         }
 110
 111         netif_tx_wake_all_queues(net);
 112
 113         nvdev = hv_get_drvdata(device_obj);
 114         rdev = nvdev->extension;
 115         if (!rdev->link_state)
 116                 netif_carrier_on(net);
 117
 118         return ret;
 119 }
 120
 121 static int netvsc_close(struct net_device *net)
 122 {
 123         struct net_device_context *net_device_ctx = netdev_priv(net);
 124         struct hv_device *device_obj = net_device_ctx->device_ctx;
 125         struct netvsc_device *nvdev = hv_get_drvdata(device_obj);
 126         int ret;
 127         u32 aread, awrite, i, msec = 10, retry = 0, retry_max = 20;
 128         struct vmbus_channel *chn;
 129
 130         netif_tx_disable(net);
 131
 132         /* Make sure netvsc_set_multicast_list doesn't re-enable filter! */
 133         cancel_work_sync(&net_device_ctx->work);
 134         ret = rndis_filter_close(device_obj);
 135         if (ret != 0) {
 136                 netdev_err(net, "unable to close device (ret %d).\n", ret);
 137                 return ret;
 138         }
 139
 140         /* Ensure pending bytes in ring are read */
 141         while (true) {
 142                 aread = 0;
 143                 for (i = 0; i < nvdev->num_chn; i++) {
 144                         chn = nvdev->chn_table[i];
 145                         if (!chn)
 146                                 continue;
 147
 148                         hv_get_ringbuffer_availbytes(&chn->inbound, &aread,
 149                                                      &awrite);
 150
 151                         if (aread)
 152                                 break;
 153
 154                         hv_get_ringbuffer_availbytes(&chn->outbound, &aread,
 155                                                      &awrite);
 156
 157                         if (aread)
 158                                 break;
 159                 }
 160
 161                 retry++;
 162                 if (retry > retry_max || aread == 0)
 163                         break;
 164
 165                 msleep(msec);
 166
 167                 if (msec < 1000)
 168                         msec *= 2;
 169         }
 170
 171         if (aread) {
 172                 netdev_err(net, "Ring buffer not empty after closing rndis\n");
 173                 ret = -ETIMEDOUT;
 174         }
 175
 176         return ret;
 177 }
 178
 179 static void *init_ppi_data(struct rndis_message *msg, u32 ppi_size,
 180                                 int pkt_type)
 181 {
 182         struct rndis_packet *rndis_pkt;
 183         struct rndis_per_packet_info *ppi;
 184
 185         rndis_pkt = &msg->msg.pkt;
 186         rndis_pkt->data_offset += ppi_size;
 187
 188         ppi = (struct rndis_per_packet_info *)((void *)rndis_pkt +
 189                 rndis_pkt->per_pkt_info_offset + rndis_pkt->per_pkt_info_len);
 190
 191         ppi->size = ppi_size;
 192         ppi->type = pkt_type;
 193         ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
 194
 195         rndis_pkt->per_pkt_info_len += ppi_size;
 196
 197         return ppi;
 198 }
 199
 200 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
 201                         void *accel_priv, select_queue_fallback_t fallback)
 202 {
 203         struct net_device_context *net_device_ctx = netdev_priv(ndev);
 204         struct hv_device *hdev =  net_device_ctx->device_ctx;
 205         struct netvsc_device *nvsc_dev = hv_get_drvdata(hdev);
 206         u32 hash;
 207         u16 q_idx = 0;
 208
 209         if (nvsc_dev == NULL || ndev->real_num_tx_queues <= 1)
 210                 return 0;
 211
 212         hash = skb_get_hash(skb);
 213         q_idx = nvsc_dev->send_table[hash % VRSS_SEND_TAB_SIZE] %
 214                 ndev->real_num_tx_queues;
 215
 216         return q_idx;
 217 }
 218
 219 void netvsc_xmit_completion(void *context)
 220 {
 221         struct hv_netvsc_packet *packet = (struct hv_netvsc_packet *)context;
 222         struct sk_buff *skb = (struct sk_buff *)
 223                 (unsigned long)packet->send_completion_tid;
 224
 225         if (skb)
 226                 dev_kfree_skb_any(skb);
 227 }
 228
 229 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
 230                         struct hv_page_buffer *pb)
 231 {
 232         int j = 0;
 233
 234         /* Deal with compund pages by ignoring unused part
 235          * of the page.
 236          */
 237         page += (offset >> PAGE_SHIFT);
 238         offset &= ~PAGE_MASK;
 239
 240         while (len > 0) {
 241                 unsigned long bytes;
 242
 243                 bytes = PAGE_SIZE - offset;
 244                 if (bytes > len)
 245                         bytes = len;
 246                 pb[j].pfn = page_to_pfn(page);
 247                 pb[j].offset = offset;
 248                 pb[j].len = bytes;
 249
 250                 offset += bytes;
 251                 len -= bytes;
 252
 253                 if (offset == PAGE_SIZE && len) {
 254                         page++;
 255                         offset = 0;
 256                         j++;
 257                 }
 258         }
 259
 260         return j + 1;
 261 }
 262
 263 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
 264                            struct hv_netvsc_packet *packet)
 265 {
 266         struct hv_page_buffer *pb = packet->page_buf;
 267         u32 slots_used = 0;
 268         char *data = skb->data;
 269         int frags = skb_shinfo(skb)->nr_frags;
 270         int i;
 271
 272         /* The packet is laid out thus:
 273          * 1. hdr: RNDIS header and PPI
 274          * 2. skb linear data
 275          * 3. skb fragment data
 276          */
 277         if (hdr != NULL)
 278                 slots_used += fill_pg_buf(virt_to_page(hdr),
 279                                         offset_in_page(hdr),
 280                                         len, &pb[slots_used]);
 281
 282         packet->rmsg_size = len;
 283         packet->rmsg_pgcnt = slots_used;
 284
 285         slots_used += fill_pg_buf(virt_to_page(data),
 286                                 offset_in_page(data),
 287                                 skb_headlen(skb), &pb[slots_used]);
 288
 289         for (i = 0; i < frags; i++) {
 290                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 291
 292                 slots_used += fill_pg_buf(skb_frag_page(frag),
 293                                         frag->page_offset,
 294                                         skb_frag_size(frag), &pb[slots_used]);
 295         }
 296         return slots_used;
 297 }
 298
 299 static int count_skb_frag_slots(struct sk_buff *skb)
 300 {
 301         int i, frags = skb_shinfo(skb)->nr_frags;
 302         int pages = 0;
 303
 304         for (i = 0; i < frags; i++) {
 305                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 306                 unsigned long size = skb_frag_size(frag);
 307                 unsigned long offset = frag->page_offset;
 308
 309                 /* Skip unused frames from start of page */
 310                 offset &= ~PAGE_MASK;
 311                 pages += PFN_UP(offset + size);
 312         }
 313         return pages;
 314 }
 315
 316 static int netvsc_get_slots(struct sk_buff *skb)
 317 {
 318         char *data = skb->data;
 319         unsigned int offset = offset_in_page(data);
 320         unsigned int len = skb_headlen(skb);
 321         int slots;
 322         int frag_slots;
 323
 324         slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
 325         frag_slots = count_skb_frag_slots(skb);
 326         return slots + frag_slots;
 327 }
 328
 329 static u32 get_net_transport_info(struct sk_buff *skb, u32 *trans_off)
 330 {
 331         u32 ret_val = TRANSPORT_INFO_NOT_IP;
 332
 333         if ((eth_hdr(skb)->h_proto != htons(ETH_P_IP)) &&
 334                 (eth_hdr(skb)->h_proto != htons(ETH_P_IPV6))) {
 335                 goto not_ip;
 336         }
 337
 338         *trans_off = skb_transport_offset(skb);
 339
 340         if ((eth_hdr(skb)->h_proto == htons(ETH_P_IP))) {
 341                 struct iphdr *iphdr = ip_hdr(skb);
 342
 343                 if (iphdr->protocol == IPPROTO_TCP)
 344                         ret_val = TRANSPORT_INFO_IPV4_TCP;
 345                 else if (iphdr->protocol == IPPROTO_UDP)
 346                         ret_val = TRANSPORT_INFO_IPV4_UDP;
 347         } else {
 348                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
 349                         ret_val = TRANSPORT_INFO_IPV6_TCP;
 350                 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
 351                         ret_val = TRANSPORT_INFO_IPV6_UDP;
 352         }
 353
 354 not_ip:
 355         return ret_val;
 356 }
 357
 358 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
 359 {
 360         struct net_device_context *net_device_ctx = netdev_priv(net);
 361         struct hv_netvsc_packet *packet = NULL;
 362         int ret;
 363         unsigned int num_data_pgs;
 364         struct rndis_message *rndis_msg;
 365         struct rndis_packet *rndis_pkt;
 366         u32 rndis_msg_size;
 367         bool isvlan;
 368         bool linear = false;
 369         struct rndis_per_packet_info *ppi;
 370         struct ndis_tcp_ip_checksum_info *csum_info;
 371         struct ndis_tcp_lso_info *lso_info;
 372         int  hdr_offset;
 373         u32 net_trans_info;
 374         u32 hash;
 375         u32 skb_length;
 376         u32 pkt_sz;
 377         struct hv_page_buffer page_buf[MAX_PAGE_BUFFER_COUNT];
 378         struct netvsc_stats *tx_stats = this_cpu_ptr(net_device_ctx->tx_stats);
 379
 380         /* We will atmost need two pages to describe the rndis
 381          * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
 382          * of pages in a single packet. If skb is scattered around
 383          * more pages we try linearizing it.
 384          */
 385
 386 check_size:
 387         skb_length = skb->len;
 388         num_data_pgs = netvsc_get_slots(skb) + 2;
 389         if (num_data_pgs > MAX_PAGE_BUFFER_COUNT && linear) {
 390                 net_alert_ratelimited("packet too big: %u pages (%u bytes)\n",
 391                                       num_data_pgs, skb->len);
 392                 ret = -EFAULT;
 393                 goto drop;
 394         } else if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
 395                 if (skb_linearize(skb)) {
 396                         net_alert_ratelimited("failed to linearize skb\n");
 397                         ret = -ENOMEM;
 398                         goto drop;
 399                 }
 400                 linear = true;
 401                 goto check_size;
 402         }
 403
 404         pkt_sz = sizeof(struct hv_netvsc_packet) + RNDIS_AND_PPI_SIZE;
 405
 406         ret = skb_cow_head(skb, pkt_sz);
 407         if (ret) {
 408                 netdev_err(net, "unable to alloc hv_netvsc_packet\n");
 409                 ret = -ENOMEM;
 410                 goto drop;
 411         }
 412         /* Use the headroom for building up the packet */
 413         packet = (struct hv_netvsc_packet *)skb->head;
 414
 415         packet->status = 0;
 416         packet->xmit_more = skb->xmit_more;
 417
 418         packet->vlan_tci = skb->vlan_tci;
 419         packet->page_buf = page_buf;
 420
 421         packet->q_idx = skb_get_queue_mapping(skb);
 422
 423         packet->is_data_pkt = true;
 424         packet->total_data_buflen = skb->len;
 425
 426         packet->rndis_msg = (struct rndis_message *)((unsigned long)packet +
 427                                 sizeof(struct hv_netvsc_packet));
 428
 429         memset(packet->rndis_msg, 0, RNDIS_AND_PPI_SIZE);
 430
 431         /* Set the completion routine */
 432         packet->send_completion = netvsc_xmit_completion;
 433         packet->send_completion_ctx = packet;
 434         packet->send_completion_tid = (unsigned long)skb;
 435
 436         isvlan = packet->vlan_tci & VLAN_TAG_PRESENT;
 437
 438         /* Add the rndis header */
 439         rndis_msg = packet->rndis_msg;
 440         rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
 441         rndis_msg->msg_len = packet->total_data_buflen;
 442         rndis_pkt = &rndis_msg->msg.pkt;
 443         rndis_pkt->data_offset = sizeof(struct rndis_packet);
 444         rndis_pkt->data_len = packet->total_data_buflen;
 445         rndis_pkt->per_pkt_info_offset = sizeof(struct rndis_packet);
 446
 447         rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
 448
 449         hash = skb_get_hash_raw(skb);
 450         if (hash != 0 && net->real_num_tx_queues > 1) {
 451                 rndis_msg_size += NDIS_HASH_PPI_SIZE;
 452                 ppi = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
 453                                     NBL_HASH_VALUE);
 454                 *(u32 *)((void *)ppi + ppi->ppi_offset) = hash;
 455         }
 456
 457         if (isvlan) {
 458                 struct ndis_pkt_8021q_info *vlan;
 459
 460                 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
 461                 ppi = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
 462                                         IEEE_8021Q_INFO);
 463                 vlan = (struct ndis_pkt_8021q_info *)((void *)ppi +
 464                                                 ppi->ppi_offset);
 465                 vlan->vlanid = packet->vlan_tci & VLAN_VID_MASK;
 466                 vlan->pri = (packet->vlan_tci & VLAN_PRIO_MASK) >>
 467                                 VLAN_PRIO_SHIFT;
 468         }
 469
 470         net_trans_info = get_net_transport_info(skb, &hdr_offset);
 471         if (net_trans_info == TRANSPORT_INFO_NOT_IP)
 472                 goto do_send;
 473
 474         /*
 475          * Setup the sendside checksum offload only if this is not a
 476          * GSO packet.
 477          */
 478         if (skb_is_gso(skb))
 479                 goto do_lso;
 480
 481         if ((skb->ip_summed == CHECKSUM_NONE) ||
 482             (skb->ip_summed == CHECKSUM_UNNECESSARY))
 483                 goto do_send;
 484
 485         rndis_msg_size += NDIS_CSUM_PPI_SIZE;
 486         ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
 487                             TCPIP_CHKSUM_PKTINFO);
 488
 489         csum_info = (struct ndis_tcp_ip_checksum_info *)((void *)ppi +
 490                         ppi->ppi_offset);
 491
 492         if (net_trans_info & (INFO_IPV4 << 16))
 493                 csum_info->transmit.is_ipv4 = 1;
 494         else
 495                 csum_info->transmit.is_ipv6 = 1;
 496
 497         if (net_trans_info & INFO_TCP) {
 498                 csum_info->transmit.tcp_checksum = 1;
 499                 csum_info->transmit.tcp_header_offset = hdr_offset;
 500         } else if (net_trans_info & INFO_UDP) {
 501                 /* UDP checksum offload is not supported on ws2008r2.
 502                  * Furthermore, on ws2012 and ws2012r2, there are some
 503                  * issues with udp checksum offload from Linux guests.
 504                  * (these are host issues).
 505                  * For now compute the checksum here.
 506                  */
 507                 struct udphdr *uh;
 508                 u16 udp_len;
 509
 510                 ret = skb_cow_head(skb, 0);
 511                 if (ret)
 512                         goto drop;
 513
 514                 uh = udp_hdr(skb);
 515                 udp_len = ntohs(uh->len);
 516                 uh->check = 0;
 517                 uh->check = csum_tcpudp_magic(ip_hdr(skb)->saddr,
 518                                               ip_hdr(skb)->daddr,
 519                                               udp_len, IPPROTO_UDP,
 520                                               csum_partial(uh, udp_len, 0));
 521                 if (uh->check == 0)
 522                         uh->check = CSUM_MANGLED_0;
 523
 524                 csum_info->transmit.udp_checksum = 0;
 525         }
 526         goto do_send;
 527
 528 do_lso:
 529         rndis_msg_size += NDIS_LSO_PPI_SIZE;
 530         ppi = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
 531                             TCP_LARGESEND_PKTINFO);
 532
 533         lso_info = (struct ndis_tcp_lso_info *)((void *)ppi +
 534                         ppi->ppi_offset);
 535
 536         lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
 537         if (net_trans_info & (INFO_IPV4 << 16)) {
 538                 lso_info->lso_v2_transmit.ip_version =
 539                         NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
 540                 ip_hdr(skb)->tot_len = 0;
 541                 ip_hdr(skb)->check = 0;
 542                 tcp_hdr(skb)->check =
 543                 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
 544                                    ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 545         } else {
 546                 lso_info->lso_v2_transmit.ip_version =
 547                         NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
 548                 ipv6_hdr(skb)->payload_len = 0;
 549                 tcp_hdr(skb)->check =
 550                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 551                                 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 552         }
 553         lso_info->lso_v2_transmit.tcp_header_offset = hdr_offset;
 554         lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
 555
 556 do_send:
 557         /* Start filling in the page buffers with the rndis hdr */
 558         rndis_msg->msg_len += rndis_msg_size;
 559         packet->total_data_buflen = rndis_msg->msg_len;
 560         packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
 561                                                skb, packet);
 562
 563         ret = netvsc_send(net_device_ctx->device_ctx, packet);
 564
 565 drop:
 566         if (ret == 0) {
 567                 u64_stats_update_begin(&tx_stats->syncp);
 568                 tx_stats->packets++;
 569                 tx_stats->bytes += skb_length;
 570                 u64_stats_update_end(&tx_stats->syncp);
 571         } else {
 572                 if (ret != -EAGAIN) {
 573                         dev_kfree_skb_any(skb);
 574                         net->stats.tx_dropped++;
 575                 }
 576         }
 577
 578         return (ret == -EAGAIN) ? NETDEV_TX_BUSY : NETDEV_TX_OK;
 579 }
 580
 581 /*
 582  * netvsc_linkstatus_callback - Link up/down notification
 583  */
 584 void netvsc_linkstatus_callback(struct hv_device *device_obj,
 585                                 struct rndis_message *resp)
 586 {
 587         struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
 588         struct net_device *net;
 589         struct net_device_context *ndev_ctx;
 590         struct netvsc_device *net_device;
 591         struct rndis_device *rdev;
 592
 593         net_device = hv_get_drvdata(device_obj);
 594         rdev = net_device->extension;
 595
 596         switch (indicate->status) {
 597         case RNDIS_STATUS_MEDIA_CONNECT:
 598                 rdev->link_state = false;
 599                 break;
 600         case RNDIS_STATUS_MEDIA_DISCONNECT:
 601                 rdev->link_state = true;
 602                 break;
 603         case RNDIS_STATUS_NETWORK_CHANGE:
 604                 rdev->link_change = true;
 605                 break;
 606         default:
 607                 return;
 608         }
 609
 610         net = net_device->ndev;
 611
 612         if (!net || net->reg_state != NETREG_REGISTERED)
 613                 return;
 614
 615         ndev_ctx = netdev_priv(net);
 616         if (!rdev->link_state) {
 617                 schedule_delayed_work(&ndev_ctx->dwork, 0);
 618                 schedule_delayed_work(&ndev_ctx->dwork, msecs_to_jiffies(20));
 619         } else {
 620                 schedule_delayed_work(&ndev_ctx->dwork, 0);
 621         }
 622 }
 623
 624 /*
 625  * netvsc_recv_callback -  Callback when we receive a packet from the
 626  * "wire" on the specified device.
 627  */
 628 int netvsc_recv_callback(struct hv_device *device_obj,
 629                                 struct hv_netvsc_packet *packet,
 630                                 struct ndis_tcp_ip_checksum_info *csum_info)
 631 {
 632         struct net_device *net;
 633         struct net_device_context *net_device_ctx;
 634         struct sk_buff *skb;
 635         struct netvsc_stats *rx_stats;
 636
 637         net = ((struct netvsc_device *)hv_get_drvdata(device_obj))->ndev;
 638         if (!net || net->reg_state != NETREG_REGISTERED) {
 639                 packet->status = NVSP_STAT_FAIL;
 640                 return 0;
 641         }
 642         net_device_ctx = netdev_priv(net);
 643         rx_stats = this_cpu_ptr(net_device_ctx->rx_stats);
 644
 645         /* Allocate a skb - TODO direct I/O to pages? */
 646         skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
 647         if (unlikely(!skb)) {
 648                 ++net->stats.rx_dropped;
 649                 packet->status = NVSP_STAT_FAIL;
 650                 return 0;
 651         }
 652
 653         /*
 654          * Copy to skb. This copy is needed here since the memory pointed by
 655          * hv_netvsc_packet cannot be deallocated
 656          */
 657         memcpy(skb_put(skb, packet->total_data_buflen), packet->data,
 658                 packet->total_data_buflen);
 659
 660         skb->protocol = eth_type_trans(skb, net);
 661         if (csum_info) {
 662                 /* We only look at the IP checksum here.
 663                  * Should we be dropping the packet if checksum
 664                  * failed? How do we deal with other checksums - TCP/UDP?
 665                  */
 666                 if (csum_info->receive.ip_checksum_succeeded)
 667                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 668                 else
 669                         skb->ip_summed = CHECKSUM_NONE;
 670         }
 671
 672         if (packet->vlan_tci & VLAN_TAG_PRESENT)
 673                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 674                                        packet->vlan_tci);
 675
 676         skb_record_rx_queue(skb, packet->channel->
 677                             offermsg.offer.sub_channel_index);
 678
 679         u64_stats_update_begin(&rx_stats->syncp);
 680         rx_stats->packets++;
 681         rx_stats->bytes += packet->total_data_buflen;
 682         u64_stats_update_end(&rx_stats->syncp);
 683
 684         /*
 685          * Pass the skb back up. Network stack will deallocate the skb when it
 686          * is done.
 687          * TODO - use NAPI?
 688          */
 689         netif_rx(skb);
 690
 691         return 0;
 692 }
 693
 694 static void netvsc_get_drvinfo(struct net_device *net,
 695                                struct ethtool_drvinfo *info)
 696 {
 697         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
 698         strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
 699 }
 700
 701 static void netvsc_get_channels(struct net_device *net,
 702                                 struct ethtool_channels *channel)
 703 {
 704         struct net_device_context *net_device_ctx = netdev_priv(net);
 705         struct hv_device *dev = net_device_ctx->device_ctx;
 706         struct netvsc_device *nvdev = hv_get_drvdata(dev);
 707
 708         if (nvdev) {
 709                 channel->max_combined   = nvdev->max_chn;
 710                 channel->combined_count = nvdev->num_chn;
 711         }
 712 }
 713
 714 static int netvsc_set_channels(struct net_device *net,
 715                                struct ethtool_channels *channels)
 716 {
 717         struct net_device_context *net_device_ctx = netdev_priv(net);
 718         struct hv_device *dev = net_device_ctx->device_ctx;
 719         struct netvsc_device *nvdev = hv_get_drvdata(dev);
 720         struct netvsc_device_info device_info;
 721         u32 num_chn;
 722         u32 max_chn;
 723         int ret = 0;
 724         bool recovering = false;
 725
 726         if (!nvdev || nvdev->destroy)
 727                 return -ENODEV;
 728
 729         num_chn = nvdev->num_chn;
 730         max_chn = min_t(u32, nvdev->max_chn, num_online_cpus());
 731
 732         if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5) {
 733                 pr_info("vRSS unsupported before NVSP Version 5\n");
 734                 return -EINVAL;
 735         }
 736
 737         /* We do not support rx, tx, or other */
 738         if (!channels ||
 739             channels->rx_count ||
 740             channels->tx_count ||
 741             channels->other_count ||
 742             (channels->combined_count < 1))
 743                 return -EINVAL;
 744
 745         if (channels->combined_count > max_chn) {
 746                 pr_info("combined channels too high, using %d\n", max_chn);
 747                 channels->combined_count = max_chn;
 748         }
 749
 750         ret = netvsc_close(net);
 751         if (ret)
 752                 goto out;
 753
 754  do_set:
 755         nvdev->start_remove = true;
 756         rndis_filter_device_remove(dev);
 757
 758         nvdev->num_chn = channels->combined_count;
 759
 760         net_device_ctx->device_ctx = dev;
 761         hv_set_drvdata(dev, net);
 762
 763         memset(&device_info, 0, sizeof(device_info));
 764         device_info.num_chn = nvdev->num_chn; /* passed to RNDIS */
 765         device_info.ring_size = ring_size;
 766         device_info.max_num_vrss_chns = max_num_vrss_chns;
 767
 768         ret = rndis_filter_device_add(dev, &device_info);
 769         if (ret) {
 770                 if (recovering) {
 771                         netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
 772                         return ret;
 773                 }
 774                 goto recover;
 775         }
 776
 777         nvdev = hv_get_drvdata(dev);
 778
 779         ret = netif_set_real_num_tx_queues(net, nvdev->num_chn);
 780         if (ret) {
 781                 if (recovering) {
 782                         netdev_err(net, "could not set tx queue count (ret %d)\n", ret);
 783                         return ret;
 784                 }
 785                 goto recover;
 786         }
 787
 788         ret = netif_set_real_num_rx_queues(net, nvdev->num_chn);
 789         if (ret) {
 790                 if (recovering) {
 791                         netdev_err(net, "could not set rx queue count (ret %d)\n", ret);
 792                         return ret;
 793                 }
 794                 goto recover;
 795         }
 796
 797  out:
 798         netvsc_open(net);
 799
 800         return ret;
 801
 802  recover:
 803         /* If the above failed, we attempt to recover through the same
 804          * process but with the original number of channels.
 805          */
 806         netdev_err(net, "could not set channels, recovering\n");
 807         recovering = true;
 808         channels->combined_count = num_chn;
 809         goto do_set;
 810 }
 811
 812 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
 813 {
 814         struct net_device_context *ndevctx = netdev_priv(ndev);
 815         struct hv_device *hdev =  ndevctx->device_ctx;
 816         struct netvsc_device *nvdev = hv_get_drvdata(hdev);
 817         struct netvsc_device_info device_info;
 818         int limit = ETH_DATA_LEN;
 819         int ret = 0;
 820
 821         if (nvdev == NULL || nvdev->destroy)
 822                 return -ENODEV;
 823
 824         if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
 825                 limit = NETVSC_MTU - ETH_HLEN;
 826
 827         if (mtu < NETVSC_MTU_MIN || mtu > limit)
 828                 return -EINVAL;
 829
 830         ret = netvsc_close(ndev);
 831         if (ret)
 832                 goto out;
 833
 834         nvdev->start_remove = true;
 835         rndis_filter_device_remove(hdev);
 836
 837         ndev->mtu = mtu;
 838
 839         ndevctx->device_ctx = hdev;
 840         hv_set_drvdata(hdev, ndev);
 841
 842         memset(&device_info, 0, sizeof(device_info));
 843         device_info.ring_size = ring_size;
 844         device_info.num_chn = nvdev->num_chn;
 845         device_info.max_num_vrss_chns = max_num_vrss_chns;
 846         rndis_filter_device_add(hdev, &device_info);
 847
 848 out:
 849         netvsc_open(ndev);
 850
 851         return ret;
 852 }
 853
 854 static struct rtnl_link_stats64 *netvsc_get_stats64(struct net_device *net,
 855                                                     struct rtnl_link_stats64 *t)
 856 {
 857         struct net_device_context *ndev_ctx = netdev_priv(net);
 858         int cpu;
 859
 860         for_each_possible_cpu(cpu) {
 861                 struct netvsc_stats *tx_stats = per_cpu_ptr(ndev_ctx->tx_stats,
 862                                                             cpu);
 863                 struct netvsc_stats *rx_stats = per_cpu_ptr(ndev_ctx->rx_stats,
 864                                                             cpu);
 865                 u64 tx_packets, tx_bytes, rx_packets, rx_bytes;
 866                 unsigned int start;
 867
 868                 do {
 869                         start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
 870                         tx_packets = tx_stats->packets;
 871                         tx_bytes = tx_stats->bytes;
 872                 } while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
 873
 874                 do {
 875                         start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
 876                         rx_packets = rx_stats->packets;
 877                         rx_bytes = rx_stats->bytes;
 878                 } while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
 879
 880                 t->tx_bytes     += tx_bytes;
 881                 t->tx_packets   += tx_packets;
 882                 t->rx_bytes     += rx_bytes;
 883                 t->rx_packets   += rx_packets;
 884         }
 885
 886         t->tx_dropped   = net->stats.tx_dropped;
 887         t->tx_errors    = net->stats.tx_dropped;
 888
 889         t->rx_dropped   = net->stats.rx_dropped;
 890         t->rx_errors    = net->stats.rx_errors;
 891
 892         return t;
 893 }
 894
 895 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
 896 {
 897         struct net_device_context *ndevctx = netdev_priv(ndev);
 898         struct hv_device *hdev =  ndevctx->device_ctx;
 899         struct sockaddr *addr = p;
 900         char save_adr[ETH_ALEN];
 901         unsigned char save_aatype;
 902         int err;
 903
 904         memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
 905         save_aatype = ndev->addr_assign_type;
 906
 907         err = eth_mac_addr(ndev, p);
 908         if (err != 0)
 909                 return err;
 910
 911         err = rndis_filter_set_device_mac(hdev, addr->sa_data);
 912         if (err != 0) {
 913                 /* roll back to saved MAC */
 914                 memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
 915                 ndev->addr_assign_type = save_aatype;
 916         }
 917
 918         return err;
 919 }
 920
 921 #ifdef CONFIG_NET_POLL_CONTROLLER
 922 static void netvsc_poll_controller(struct net_device *net)
 923 {
 924         /* As netvsc_start_xmit() works synchronous we don't have to
 925          * trigger anything here.
 926          */
 927 }
 928 #endif
 929
 930 static const struct ethtool_ops ethtool_ops = {
 931         .get_drvinfo    = netvsc_get_drvinfo,
 932         .get_link       = ethtool_op_get_link,
 933         .get_channels   = netvsc_get_channels,
 934         .set_channels   = netvsc_set_channels,
 935 };
 936
 937 static const struct net_device_ops device_ops = {
 938         .ndo_open =                     netvsc_open,
 939         .ndo_stop =                     netvsc_close,
 940         .ndo_start_xmit =               netvsc_start_xmit,
 941         .ndo_set_rx_mode =              netvsc_set_multicast_list,
 942         .ndo_change_mtu =               netvsc_change_mtu,
 943         .ndo_validate_addr =            eth_validate_addr,
 944         .ndo_set_mac_address =          netvsc_set_mac_addr,
 945         .ndo_select_queue =             netvsc_select_queue,
 946         .ndo_get_stats64 =              netvsc_get_stats64,
 947 #ifdef CONFIG_NET_POLL_CONTROLLER
 948         .ndo_poll_controller =          netvsc_poll_controller,
 949 #endif
 950 };
 951
 952 /*
 953  * Send GARP packet to network peers after migrations.
 954  * After Quick Migration, the network is not immediately operational in the
 955  * current context when receiving RNDIS_STATUS_MEDIA_CONNECT event. So, add
 956  * another netif_notify_peers() into a delayed work, otherwise GARP packet
 957  * will not be sent after quick migration, and cause network disconnection.
 958  * Also, we update the carrier status here.
 959  */
 960 static void netvsc_link_change(struct work_struct *w)
 961 {
 962         struct net_device_context *ndev_ctx;
 963         struct net_device *net;
 964         struct netvsc_device *net_device;
 965         struct rndis_device *rdev;
 966         bool notify, refresh = false;
 967         char *argv[] = { "/etc/init.d/network", "restart", NULL };
 968         char *envp[] = { "HOME=/", "PATH=/sbin:/usr/sbin:/bin:/usr/bin", NULL };
 969
 970         rtnl_lock();
 971
 972         ndev_ctx = container_of(w, struct net_device_context, dwork.work);
 973         net_device = hv_get_drvdata(ndev_ctx->device_ctx);
 974         rdev = net_device->extension;
 975         net = net_device->ndev;
 976
 977         if (rdev->link_state) {
 978                 netif_carrier_off(net);
 979                 notify = false;
 980         } else {
 981                 netif_carrier_on(net);
 982                 notify = true;
 983                 if (rdev->link_change) {
 984                         rdev->link_change = false;
 985                         refresh = true;
 986                 }
 987         }
 988
 989         rtnl_unlock();
 990
 991         if (refresh)
 992                 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
 993
 994         if (notify)
 995                 netdev_notify_peers(net);
 996 }
 997
 998 static void netvsc_free_netdev(struct net_device *netdev)
 999 {
1000         struct net_device_context *net_device_ctx = netdev_priv(netdev);
1001
1002         free_percpu(net_device_ctx->tx_stats);
1003         free_percpu(net_device_ctx->rx_stats);
1004         free_netdev(netdev);
1005 }
1006
1007 static int netvsc_probe(struct hv_device *dev,
1008                         const struct hv_vmbus_device_id *dev_id)
1009 {
1010         struct net_device *net = NULL;
1011         struct net_device_context *net_device_ctx;
1012         struct netvsc_device_info device_info;
1013         struct netvsc_device *nvdev;
1014         int ret;
1015         u32 max_needed_headroom;
1016
1017         net = alloc_etherdev_mq(sizeof(struct net_device_context),
1018                                 num_online_cpus());
1019         if (!net)
1020                 return -ENOMEM;
1021
1022         max_needed_headroom = sizeof(struct hv_netvsc_packet) +
1023                               RNDIS_AND_PPI_SIZE;
1024
1025         netif_carrier_off(net);
1026
1027         net_device_ctx = netdev_priv(net);
1028         net_device_ctx->device_ctx = dev;
1029         net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
1030         if (netif_msg_probe(net_device_ctx))
1031                 netdev_dbg(net, "netvsc msg_enable: %d\n",
1032                            net_device_ctx->msg_enable);
1033
1034         net_device_ctx->tx_stats = netdev_alloc_pcpu_stats(struct netvsc_stats);
1035         if (!net_device_ctx->tx_stats) {
1036                 free_netdev(net);
1037                 return -ENOMEM;
1038         }
1039         net_device_ctx->rx_stats = netdev_alloc_pcpu_stats(struct netvsc_stats);
1040         if (!net_device_ctx->rx_stats) {
1041                 free_percpu(net_device_ctx->tx_stats);
1042                 free_netdev(net);
1043                 return -ENOMEM;
1044         }
1045
1046         hv_set_drvdata(dev, net);
1047         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
1048         INIT_WORK(&net_device_ctx->work, do_set_multicast);
1049
1050         net->netdev_ops = &device_ops;
1051
1052         net->hw_features = NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_IP_CSUM |
1053                                 NETIF_F_TSO;
1054         net->features = NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_SG | NETIF_F_RXCSUM |
1055                         NETIF_F_IP_CSUM | NETIF_F_TSO;
1056
1057         net->ethtool_ops = &ethtool_ops;
1058         SET_NETDEV_DEV(net, &dev->device);
1059
1060         /*
1061          * Request additional head room in the skb.
1062          * We will use this space to build the rndis
1063          * heaser and other state we need to maintain.
1064          */
1065         net->needed_headroom = max_needed_headroom;
1066
1067         /* Notify the netvsc driver of the new device */
1068         memset(&device_info, 0, sizeof(device_info));
1069         device_info.ring_size = ring_size;
1070         device_info.max_num_vrss_chns = max_num_vrss_chns;
1071         ret = rndis_filter_device_add(dev, &device_info);
1072         if (ret != 0) {
1073                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
1074                 netvsc_free_netdev(net);
1075                 hv_set_drvdata(dev, NULL);
1076                 return ret;
1077         }
1078         memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
1079
1080         nvdev = hv_get_drvdata(dev);
1081         netif_set_real_num_tx_queues(net, nvdev->num_chn);
1082         netif_set_real_num_rx_queues(net, nvdev->num_chn);
1083         netif_set_gso_max_size(net, NETVSC_GSO_MAX_SIZE);
1084
1085         ret = register_netdev(net);
1086         if (ret != 0) {
1087                 pr_err("Unable to register netdev.\n");
1088                 rndis_filter_device_remove(dev);
1089                 netvsc_free_netdev(net);
1090         } else {
1091                 schedule_delayed_work(&net_device_ctx->dwork, 0);
1092         }
1093
1094         return ret;
1095 }
1096
1097 static int netvsc_remove(struct hv_device *dev)
1098 {
1099         struct net_device *net;
1100         struct net_device_context *ndev_ctx;
1101         struct netvsc_device *net_device;
1102
1103         net_device = hv_get_drvdata(dev);
1104         net = net_device->ndev;
1105
1106         if (net == NULL) {
1107                 dev_err(&dev->device, "No net device to remove\n");
1108                 return 0;
1109         }
1110
1111         net_device->start_remove = true;
1112
1113         ndev_ctx = netdev_priv(net);
1114         cancel_delayed_work_sync(&ndev_ctx->dwork);
1115         cancel_work_sync(&ndev_ctx->work);
1116
1117         /* Stop outbound asap */
1118         netif_tx_disable(net);
1119
1120         unregister_netdev(net);
1121
1122         /*
1123          * Call to the vsc driver to let it know that the device is being
1124          * removed
1125          */
1126         rndis_filter_device_remove(dev);
1127
1128         netvsc_free_netdev(net);
1129         return 0;
1130 }
1131
1132 static const struct hv_vmbus_device_id id_table[] = {
1133         /* Network guid */
1134         { HV_NIC_GUID, },
1135         { },
1136 };
1137
1138 MODULE_DEVICE_TABLE(vmbus, id_table);
1139
1140 /* The one and only one */
1141 static struct  hv_driver netvsc_drv = {
1142         .name = KBUILD_MODNAME,
1143         .id_table = id_table,
1144         .probe = netvsc_probe,
1145         .remove = netvsc_remove,
1146 };
1147
1148 static void __exit netvsc_drv_exit(void)
1149 {
1150         vmbus_driver_unregister(&netvsc_drv);
1151 }
1152
1153 static int __init netvsc_drv_init(void)
1154 {
1155         if (ring_size < RING_SIZE_MIN) {
1156                 ring_size = RING_SIZE_MIN;
1157                 pr_info("Increased ring_size to %d (min allowed)\n",
1158                         ring_size);
1159         }
1160         return vmbus_driver_register(&netvsc_drv);
1161 }
1162
1163 MODULE_LICENSE("GPL");
1164 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
1165
1166 module_init(netvsc_drv_init);
1167 module_exit(netvsc_drv_exit);