drivers/net/hyperv/netvsc_drv.c

   1 /*
   2  * Copyright (c) 2009, Microsoft Corporation.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms and conditions of the GNU General Public License,
   6  * version 2, as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope it will be useful, but WITHOUT
   9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  11  * more details.
  12  *
  13  * You should have received a copy of the GNU General Public License along with
  14  * this program; if not, see <http://www.gnu.org/licenses/>.
  15  *
  16  * Authors:
  17  *   Haiyang Zhang <haiyangz@microsoft.com>
  18  *   Hank Janssen  <hjanssen@microsoft.com>
  19  */
  20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  21
  22 #include <linux/init.h>
  23 #include <linux/atomic.h>
  24 #include <linux/module.h>
  25 #include <linux/highmem.h>
  26 #include <linux/device.h>
  27 #include <linux/io.h>
  28 #include <linux/delay.h>
  29 #include <linux/netdevice.h>
  30 #include <linux/inetdevice.h>
  31 #include <linux/etherdevice.h>
  32 #include <linux/skbuff.h>
  33 #include <linux/if_vlan.h>
  34 #include <linux/in.h>
  35 #include <linux/slab.h>
  36 #include <net/arp.h>
  37 #include <net/route.h>
  38 #include <net/sock.h>
  39 #include <net/pkt_sched.h>
  40
  41 #include "hyperv_net.h"
  42
  43
  44 #define RING_SIZE_MIN 64
  45 #define LINKCHANGE_INT (2 * HZ)
  46 static int ring_size = 128;
  47 module_param(ring_size, int, S_IRUGO);
  48 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
  49
  50 static int max_num_vrss_chns = 8;
  51
  52 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
  53                                 NETIF_MSG_LINK | NETIF_MSG_IFUP |
  54                                 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
  55                                 NETIF_MSG_TX_ERR;
  56
  57 static int debug = -1;
  58 module_param(debug, int, S_IRUGO);
  59 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  60
  61 static void do_set_multicast(struct work_struct *w)
  62 {
  63         struct net_device_context *ndevctx =
  64                 container_of(w, struct net_device_context, work);
  65         struct netvsc_device *nvdev;
  66         struct rndis_device *rdev;
  67
  68         nvdev = hv_get_drvdata(ndevctx->device_ctx);
  69         if (nvdev == NULL || nvdev->ndev == NULL)
  70                 return;
  71
  72         rdev = nvdev->extension;
  73         if (rdev == NULL)
  74                 return;
  75
  76         if (nvdev->ndev->flags & IFF_PROMISC)
  77                 rndis_filter_set_packet_filter(rdev,
  78                         NDIS_PACKET_TYPE_PROMISCUOUS);
  79         else
  80                 rndis_filter_set_packet_filter(rdev,
  81                         NDIS_PACKET_TYPE_BROADCAST |
  82                         NDIS_PACKET_TYPE_ALL_MULTICAST |
  83                         NDIS_PACKET_TYPE_DIRECTED);
  84 }
  85
  86 static void netvsc_set_multicast_list(struct net_device *net)
  87 {
  88         struct net_device_context *net_device_ctx = netdev_priv(net);
  89
  90         schedule_work(&net_device_ctx->work);
  91 }
  92
  93 static int netvsc_open(struct net_device *net)
  94 {
  95         struct net_device_context *net_device_ctx = netdev_priv(net);
  96         struct hv_device *device_obj = net_device_ctx->device_ctx;
  97         struct netvsc_device *nvdev;
  98         struct rndis_device *rdev;
  99         int ret = 0;
 100
 101         netif_carrier_off(net);
 102
 103         /* Open up the device */
 104         ret = rndis_filter_open(device_obj);
 105         if (ret != 0) {
 106                 netdev_err(net, "unable to open device (ret %d).\n", ret);
 107                 return ret;
 108         }
 109
 110         netif_tx_wake_all_queues(net);
 111
 112         nvdev = hv_get_drvdata(device_obj);
 113         rdev = nvdev->extension;
 114         if (!rdev->link_state)
 115                 netif_carrier_on(net);
 116
 117         return ret;
 118 }
 119
 120 static int netvsc_close(struct net_device *net)
 121 {
 122         struct net_device_context *net_device_ctx = netdev_priv(net);
 123         struct hv_device *device_obj = net_device_ctx->device_ctx;
 124         struct netvsc_device *nvdev = hv_get_drvdata(device_obj);
 125         int ret;
 126         u32 aread, awrite, i, msec = 10, retry = 0, retry_max = 20;
 127         struct vmbus_channel *chn;
 128
 129         netif_tx_disable(net);
 130
 131         /* Make sure netvsc_set_multicast_list doesn't re-enable filter! */
 132         cancel_work_sync(&net_device_ctx->work);
 133         ret = rndis_filter_close(device_obj);
 134         if (ret != 0) {
 135                 netdev_err(net, "unable to close device (ret %d).\n", ret);
 136                 return ret;
 137         }
 138
 139         /* Ensure pending bytes in ring are read */
 140         while (true) {
 141                 aread = 0;
 142                 for (i = 0; i < nvdev->num_chn; i++) {
 143                         chn = nvdev->chn_table[i];
 144                         if (!chn)
 145                                 continue;
 146
 147                         hv_get_ringbuffer_availbytes(&chn->inbound, &aread,
 148                                                      &awrite);
 149
 150                         if (aread)
 151                                 break;
 152
 153                         hv_get_ringbuffer_availbytes(&chn->outbound, &aread,
 154                                                      &awrite);
 155
 156                         if (aread)
 157                                 break;
 158                 }
 159
 160                 retry++;
 161                 if (retry > retry_max || aread == 0)
 162                         break;
 163
 164                 msleep(msec);
 165
 166                 if (msec < 1000)
 167                         msec *= 2;
 168         }
 169
 170         if (aread) {
 171                 netdev_err(net, "Ring buffer not empty after closing rndis\n");
 172                 ret = -ETIMEDOUT;
 173         }
 174
 175         return ret;
 176 }
 177
 178 static void *init_ppi_data(struct rndis_message *msg, u32 ppi_size,
 179                                 int pkt_type)
 180 {
 181         struct rndis_packet *rndis_pkt;
 182         struct rndis_per_packet_info *ppi;
 183
 184         rndis_pkt = &msg->msg.pkt;
 185         rndis_pkt->data_offset += ppi_size;
 186
 187         ppi = (struct rndis_per_packet_info *)((void *)rndis_pkt +
 188                 rndis_pkt->per_pkt_info_offset + rndis_pkt->per_pkt_info_len);
 189
 190         ppi->size = ppi_size;
 191         ppi->type = pkt_type;
 192         ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
 193
 194         rndis_pkt->per_pkt_info_len += ppi_size;
 195
 196         return ppi;
 197 }
 198
 199 union sub_key {
 200         u64 k;
 201         struct {
 202                 u8 pad[3];
 203                 u8 kb;
 204                 u32 ka;
 205         };
 206 };
 207
 208 /* Toeplitz hash function
 209  * data: network byte order
 210  * return: host byte order
 211  */
 212 static u32 comp_hash(u8 *key, int klen, void *data, int dlen)
 213 {
 214         union sub_key subk;
 215         int k_next = 4;
 216         u8 dt;
 217         int i, j;
 218         u32 ret = 0;
 219
 220         subk.k = 0;
 221         subk.ka = ntohl(*(u32 *)key);
 222
 223         for (i = 0; i < dlen; i++) {
 224                 subk.kb = key[k_next];
 225                 k_next = (k_next + 1) % klen;
 226                 dt = ((u8 *)data)[i];
 227                 for (j = 0; j < 8; j++) {
 228                         if (dt & 0x80)
 229                                 ret ^= subk.ka;
 230                         dt <<= 1;
 231                         subk.k <<= 1;
 232                 }
 233         }
 234
 235         return ret;
 236 }
 237
 238 static bool netvsc_set_hash(u32 *hash, struct sk_buff *skb)
 239 {
 240         struct flow_keys flow;
 241         int data_len;
 242
 243         if (!skb_flow_dissect_flow_keys(skb, &flow, 0) ||
 244             !(flow.basic.n_proto == htons(ETH_P_IP) ||
 245               flow.basic.n_proto == htons(ETH_P_IPV6)))
 246                 return false;
 247
 248         if (flow.basic.ip_proto == IPPROTO_TCP)
 249                 data_len = 12;
 250         else
 251                 data_len = 8;
 252
 253         *hash = comp_hash(netvsc_hash_key, HASH_KEYLEN, &flow, data_len);
 254
 255         return true;
 256 }
 257
 258 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
 259                         void *accel_priv, select_queue_fallback_t fallback)
 260 {
 261         struct net_device_context *net_device_ctx = netdev_priv(ndev);
 262         struct hv_device *hdev =  net_device_ctx->device_ctx;
 263         struct netvsc_device *nvsc_dev = hv_get_drvdata(hdev);
 264         u32 hash;
 265         u16 q_idx = 0;
 266
 267         if (nvsc_dev == NULL || ndev->real_num_tx_queues <= 1)
 268                 return 0;
 269
 270         if (netvsc_set_hash(&hash, skb)) {
 271                 q_idx = nvsc_dev->send_table[hash % VRSS_SEND_TAB_SIZE] %
 272                         ndev->real_num_tx_queues;
 273                 skb_set_hash(skb, hash, PKT_HASH_TYPE_L3);
 274         }
 275
 276         if (!nvsc_dev->chn_table[q_idx])
 277                 q_idx = 0;
 278
 279         return q_idx;
 280 }
 281
 282 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
 283                         struct hv_page_buffer *pb)
 284 {
 285         int j = 0;
 286
 287         /* Deal with compund pages by ignoring unused part
 288          * of the page.
 289          */
 290         page += (offset >> PAGE_SHIFT);
 291         offset &= ~PAGE_MASK;
 292
 293         while (len > 0) {
 294                 unsigned long bytes;
 295
 296                 bytes = PAGE_SIZE - offset;
 297                 if (bytes > len)
 298                         bytes = len;
 299                 pb[j].pfn = page_to_pfn(page);
 300                 pb[j].offset = offset;
 301                 pb[j].len = bytes;
 302
 303                 offset += bytes;
 304                 len -= bytes;
 305
 306                 if (offset == PAGE_SIZE && len) {
 307                         page++;
 308                         offset = 0;
 309                         j++;
 310                 }
 311         }
 312
 313         return j + 1;
 314 }
 315
 316 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
 317                            struct hv_netvsc_packet *packet,
 318                            struct hv_page_buffer **page_buf)
 319 {
 320         struct hv_page_buffer *pb = *page_buf;
 321         u32 slots_used = 0;
 322         char *data = skb->data;
 323         int frags = skb_shinfo(skb)->nr_frags;
 324         int i;
 325
 326         /* The packet is laid out thus:
 327          * 1. hdr: RNDIS header and PPI
 328          * 2. skb linear data
 329          * 3. skb fragment data
 330          */
 331         if (hdr != NULL)
 332                 slots_used += fill_pg_buf(virt_to_page(hdr),
 333                                         offset_in_page(hdr),
 334                                         len, &pb[slots_used]);
 335
 336         packet->rmsg_size = len;
 337         packet->rmsg_pgcnt = slots_used;
 338
 339         slots_used += fill_pg_buf(virt_to_page(data),
 340                                 offset_in_page(data),
 341                                 skb_headlen(skb), &pb[slots_used]);
 342
 343         for (i = 0; i < frags; i++) {
 344                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 345
 346                 slots_used += fill_pg_buf(skb_frag_page(frag),
 347                                         frag->page_offset,
 348                                         skb_frag_size(frag), &pb[slots_used]);
 349         }
 350         return slots_used;
 351 }
 352
 353 static int count_skb_frag_slots(struct sk_buff *skb)
 354 {
 355         int i, frags = skb_shinfo(skb)->nr_frags;
 356         int pages = 0;
 357
 358         for (i = 0; i < frags; i++) {
 359                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 360                 unsigned long size = skb_frag_size(frag);
 361                 unsigned long offset = frag->page_offset;
 362
 363                 /* Skip unused frames from start of page */
 364                 offset &= ~PAGE_MASK;
 365                 pages += PFN_UP(offset + size);
 366         }
 367         return pages;
 368 }
 369
 370 static int netvsc_get_slots(struct sk_buff *skb)
 371 {
 372         char *data = skb->data;
 373         unsigned int offset = offset_in_page(data);
 374         unsigned int len = skb_headlen(skb);
 375         int slots;
 376         int frag_slots;
 377
 378         slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
 379         frag_slots = count_skb_frag_slots(skb);
 380         return slots + frag_slots;
 381 }
 382
 383 static u32 get_net_transport_info(struct sk_buff *skb, u32 *trans_off)
 384 {
 385         u32 ret_val = TRANSPORT_INFO_NOT_IP;
 386
 387         if ((eth_hdr(skb)->h_proto != htons(ETH_P_IP)) &&
 388                 (eth_hdr(skb)->h_proto != htons(ETH_P_IPV6))) {
 389                 goto not_ip;
 390         }
 391
 392         *trans_off = skb_transport_offset(skb);
 393
 394         if ((eth_hdr(skb)->h_proto == htons(ETH_P_IP))) {
 395                 struct iphdr *iphdr = ip_hdr(skb);
 396
 397                 if (iphdr->protocol == IPPROTO_TCP)
 398                         ret_val = TRANSPORT_INFO_IPV4_TCP;
 399                 else if (iphdr->protocol == IPPROTO_UDP)
 400                         ret_val = TRANSPORT_INFO_IPV4_UDP;
 401         } else {
 402                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
 403                         ret_val = TRANSPORT_INFO_IPV6_TCP;
 404                 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
 405                         ret_val = TRANSPORT_INFO_IPV6_UDP;
 406         }
 407
 408 not_ip:
 409         return ret_val;
 410 }
 411
 412 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
 413 {
 414         struct net_device_context *net_device_ctx = netdev_priv(net);
 415         struct hv_netvsc_packet *packet = NULL;
 416         int ret;
 417         unsigned int num_data_pgs;
 418         struct rndis_message *rndis_msg;
 419         struct rndis_packet *rndis_pkt;
 420         u32 rndis_msg_size;
 421         bool isvlan;
 422         bool linear = false;
 423         struct rndis_per_packet_info *ppi;
 424         struct ndis_tcp_ip_checksum_info *csum_info;
 425         struct ndis_tcp_lso_info *lso_info;
 426         int  hdr_offset;
 427         u32 net_trans_info;
 428         u32 hash;
 429         u32 skb_length;
 430         struct hv_page_buffer page_buf[MAX_PAGE_BUFFER_COUNT];
 431         struct hv_page_buffer *pb = page_buf;
 432         struct netvsc_stats *tx_stats = this_cpu_ptr(net_device_ctx->tx_stats);
 433
 434         /* We will atmost need two pages to describe the rndis
 435          * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
 436          * of pages in a single packet. If skb is scattered around
 437          * more pages we try linearizing it.
 438          */
 439
 440 check_size:
 441         skb_length = skb->len;
 442         num_data_pgs = netvsc_get_slots(skb) + 2;
 443         if (num_data_pgs > MAX_PAGE_BUFFER_COUNT && linear) {
 444                 net_alert_ratelimited("packet too big: %u pages (%u bytes)\n",
 445                                       num_data_pgs, skb->len);
 446                 ret = -EFAULT;
 447                 goto drop;
 448         } else if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
 449                 if (skb_linearize(skb)) {
 450                         net_alert_ratelimited("failed to linearize skb\n");
 451                         ret = -ENOMEM;
 452                         goto drop;
 453                 }
 454                 linear = true;
 455                 goto check_size;
 456         }
 457
 458         /*
 459          * Place the rndis header in the skb head room and
 460          * the skb->cb will be used for hv_netvsc_packet
 461          * structure.
 462          */
 463         ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
 464         if (ret) {
 465                 netdev_err(net, "unable to alloc hv_netvsc_packet\n");
 466                 ret = -ENOMEM;
 467                 goto drop;
 468         }
 469         /* Use the skb control buffer for building up the packet */
 470         BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
 471                         FIELD_SIZEOF(struct sk_buff, cb));
 472         packet = (struct hv_netvsc_packet *)skb->cb;
 473
 474         packet->status = 0;
 475         packet->xmit_more = skb->xmit_more;
 476
 477         packet->vlan_tci = skb->vlan_tci;
 478
 479         packet->q_idx = skb_get_queue_mapping(skb);
 480
 481         packet->total_data_buflen = skb->len;
 482
 483         rndis_msg = (struct rndis_message *)skb->head;
 484
 485         memset(rndis_msg, 0, RNDIS_AND_PPI_SIZE);
 486
 487         isvlan = packet->vlan_tci & VLAN_TAG_PRESENT;
 488
 489         /* Add the rndis header */
 490         rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
 491         rndis_msg->msg_len = packet->total_data_buflen;
 492         rndis_pkt = &rndis_msg->msg.pkt;
 493         rndis_pkt->data_offset = sizeof(struct rndis_packet);
 494         rndis_pkt->data_len = packet->total_data_buflen;
 495         rndis_pkt->per_pkt_info_offset = sizeof(struct rndis_packet);
 496
 497         rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
 498
 499         hash = skb_get_hash_raw(skb);
 500         if (hash != 0 && net->real_num_tx_queues > 1) {
 501                 rndis_msg_size += NDIS_HASH_PPI_SIZE;
 502                 ppi = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
 503                                     NBL_HASH_VALUE);
 504                 *(u32 *)((void *)ppi + ppi->ppi_offset) = hash;
 505         }
 506
 507         if (isvlan) {
 508                 struct ndis_pkt_8021q_info *vlan;
 509
 510                 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
 511                 ppi = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
 512                                         IEEE_8021Q_INFO);
 513                 vlan = (struct ndis_pkt_8021q_info *)((void *)ppi +
 514                                                 ppi->ppi_offset);
 515                 vlan->vlanid = packet->vlan_tci & VLAN_VID_MASK;
 516                 vlan->pri = (packet->vlan_tci & VLAN_PRIO_MASK) >>
 517                                 VLAN_PRIO_SHIFT;
 518         }
 519
 520         net_trans_info = get_net_transport_info(skb, &hdr_offset);
 521         if (net_trans_info == TRANSPORT_INFO_NOT_IP)
 522                 goto do_send;
 523
 524         /*
 525          * Setup the sendside checksum offload only if this is not a
 526          * GSO packet.
 527          */
 528         if (skb_is_gso(skb))
 529                 goto do_lso;
 530
 531         if ((skb->ip_summed == CHECKSUM_NONE) ||
 532             (skb->ip_summed == CHECKSUM_UNNECESSARY))
 533                 goto do_send;
 534
 535         rndis_msg_size += NDIS_CSUM_PPI_SIZE;
 536         ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
 537                             TCPIP_CHKSUM_PKTINFO);
 538
 539         csum_info = (struct ndis_tcp_ip_checksum_info *)((void *)ppi +
 540                         ppi->ppi_offset);
 541
 542         if (net_trans_info & (INFO_IPV4 << 16))
 543                 csum_info->transmit.is_ipv4 = 1;
 544         else
 545                 csum_info->transmit.is_ipv6 = 1;
 546
 547         if (net_trans_info & INFO_TCP) {
 548                 csum_info->transmit.tcp_checksum = 1;
 549                 csum_info->transmit.tcp_header_offset = hdr_offset;
 550         } else if (net_trans_info & INFO_UDP) {
 551                 /* UDP checksum offload is not supported on ws2008r2.
 552                  * Furthermore, on ws2012 and ws2012r2, there are some
 553                  * issues with udp checksum offload from Linux guests.
 554                  * (these are host issues).
 555                  * For now compute the checksum here.
 556                  */
 557                 struct udphdr *uh;
 558                 u16 udp_len;
 559
 560                 ret = skb_cow_head(skb, 0);
 561                 if (ret)
 562                         goto drop;
 563
 564                 uh = udp_hdr(skb);
 565                 udp_len = ntohs(uh->len);
 566                 uh->check = 0;
 567                 uh->check = csum_tcpudp_magic(ip_hdr(skb)->saddr,
 568                                               ip_hdr(skb)->daddr,
 569                                               udp_len, IPPROTO_UDP,
 570                                               csum_partial(uh, udp_len, 0));
 571                 if (uh->check == 0)
 572                         uh->check = CSUM_MANGLED_0;
 573
 574                 csum_info->transmit.udp_checksum = 0;
 575         }
 576         goto do_send;
 577
 578 do_lso:
 579         rndis_msg_size += NDIS_LSO_PPI_SIZE;
 580         ppi = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
 581                             TCP_LARGESEND_PKTINFO);
 582
 583         lso_info = (struct ndis_tcp_lso_info *)((void *)ppi +
 584                         ppi->ppi_offset);
 585
 586         lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
 587         if (net_trans_info & (INFO_IPV4 << 16)) {
 588                 lso_info->lso_v2_transmit.ip_version =
 589                         NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
 590                 ip_hdr(skb)->tot_len = 0;
 591                 ip_hdr(skb)->check = 0;
 592                 tcp_hdr(skb)->check =
 593                 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
 594                                    ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 595         } else {
 596                 lso_info->lso_v2_transmit.ip_version =
 597                         NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
 598                 ipv6_hdr(skb)->payload_len = 0;
 599                 tcp_hdr(skb)->check =
 600                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 601                                 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 602         }
 603         lso_info->lso_v2_transmit.tcp_header_offset = hdr_offset;
 604         lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
 605
 606 do_send:
 607         /* Start filling in the page buffers with the rndis hdr */
 608         rndis_msg->msg_len += rndis_msg_size;
 609         packet->total_data_buflen = rndis_msg->msg_len;
 610         packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
 611                                                skb, packet, &pb);
 612
 613         ret = netvsc_send(net_device_ctx->device_ctx, packet,
 614                           rndis_msg, &pb, skb);
 615
 616 drop:
 617         if (ret == 0) {
 618                 u64_stats_update_begin(&tx_stats->syncp);
 619                 tx_stats->packets++;
 620                 tx_stats->bytes += skb_length;
 621                 u64_stats_update_end(&tx_stats->syncp);
 622         } else {
 623                 if (ret != -EAGAIN) {
 624                         dev_kfree_skb_any(skb);
 625                         net->stats.tx_dropped++;
 626                 }
 627         }
 628
 629         return (ret == -EAGAIN) ? NETDEV_TX_BUSY : NETDEV_TX_OK;
 630 }
 631
 632 /*
 633  * netvsc_linkstatus_callback - Link up/down notification
 634  */
 635 void netvsc_linkstatus_callback(struct hv_device *device_obj,
 636                                 struct rndis_message *resp)
 637 {
 638         struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
 639         struct net_device *net;
 640         struct net_device_context *ndev_ctx;
 641         struct netvsc_device *net_device;
 642         struct netvsc_reconfig *event;
 643         unsigned long flags;
 644
 645         /* Handle link change statuses only */
 646         if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
 647             indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
 648             indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
 649                 return;
 650
 651         net_device = hv_get_drvdata(device_obj);
 652         net = net_device->ndev;
 653
 654         if (!net || net->reg_state != NETREG_REGISTERED)
 655                 return;
 656
 657         ndev_ctx = netdev_priv(net);
 658
 659         event = kzalloc(sizeof(*event), GFP_ATOMIC);
 660         if (!event)
 661                 return;
 662         event->event = indicate->status;
 663
 664         spin_lock_irqsave(&ndev_ctx->lock, flags);
 665         list_add_tail(&event->list, &ndev_ctx->reconfig_events);
 666         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
 667
 668         schedule_delayed_work(&ndev_ctx->dwork, 0);
 669 }
 670
 671 /*
 672  * netvsc_recv_callback -  Callback when we receive a packet from the
 673  * "wire" on the specified device.
 674  */
 675 int netvsc_recv_callback(struct hv_device *device_obj,
 676                                 struct hv_netvsc_packet *packet,
 677                                 void **data,
 678                                 struct ndis_tcp_ip_checksum_info *csum_info,
 679                                 struct vmbus_channel *channel)
 680 {
 681         struct net_device *net;
 682         struct net_device_context *net_device_ctx;
 683         struct sk_buff *skb;
 684         struct netvsc_stats *rx_stats;
 685
 686         net = ((struct netvsc_device *)hv_get_drvdata(device_obj))->ndev;
 687         if (!net || net->reg_state != NETREG_REGISTERED) {
 688                 packet->status = NVSP_STAT_FAIL;
 689                 return 0;
 690         }
 691         net_device_ctx = netdev_priv(net);
 692         rx_stats = this_cpu_ptr(net_device_ctx->rx_stats);
 693
 694         /* Allocate a skb - TODO direct I/O to pages? */
 695         skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
 696         if (unlikely(!skb)) {
 697                 ++net->stats.rx_dropped;
 698                 packet->status = NVSP_STAT_FAIL;
 699                 return 0;
 700         }
 701
 702         /*
 703          * Copy to skb. This copy is needed here since the memory pointed by
 704          * hv_netvsc_packet cannot be deallocated
 705          */
 706         memcpy(skb_put(skb, packet->total_data_buflen), *data,
 707                 packet->total_data_buflen);
 708
 709         skb->protocol = eth_type_trans(skb, net);
 710         if (csum_info) {
 711                 /* We only look at the IP checksum here.
 712                  * Should we be dropping the packet if checksum
 713                  * failed? How do we deal with other checksums - TCP/UDP?
 714                  */
 715                 if (csum_info->receive.ip_checksum_succeeded)
 716                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 717                 else
 718                         skb->ip_summed = CHECKSUM_NONE;
 719         }
 720
 721         if (packet->vlan_tci & VLAN_TAG_PRESENT)
 722                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 723                                        packet->vlan_tci);
 724
 725         skb_record_rx_queue(skb, channel->
 726                             offermsg.offer.sub_channel_index);
 727
 728         u64_stats_update_begin(&rx_stats->syncp);
 729         rx_stats->packets++;
 730         rx_stats->bytes += packet->total_data_buflen;
 731         u64_stats_update_end(&rx_stats->syncp);
 732
 733         /*
 734          * Pass the skb back up. Network stack will deallocate the skb when it
 735          * is done.
 736          * TODO - use NAPI?
 737          */
 738         netif_rx(skb);
 739
 740         return 0;
 741 }
 742
 743 static void netvsc_get_drvinfo(struct net_device *net,
 744                                struct ethtool_drvinfo *info)
 745 {
 746         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
 747         strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
 748 }
 749
 750 static void netvsc_get_channels(struct net_device *net,
 751                                 struct ethtool_channels *channel)
 752 {
 753         struct net_device_context *net_device_ctx = netdev_priv(net);
 754         struct hv_device *dev = net_device_ctx->device_ctx;
 755         struct netvsc_device *nvdev = hv_get_drvdata(dev);
 756
 757         if (nvdev) {
 758                 channel->max_combined   = nvdev->max_chn;
 759                 channel->combined_count = nvdev->num_chn;
 760         }
 761 }
 762
 763 static int netvsc_set_channels(struct net_device *net,
 764                                struct ethtool_channels *channels)
 765 {
 766         struct net_device_context *net_device_ctx = netdev_priv(net);
 767         struct hv_device *dev = net_device_ctx->device_ctx;
 768         struct netvsc_device *nvdev = hv_get_drvdata(dev);
 769         struct netvsc_device_info device_info;
 770         u32 num_chn;
 771         u32 max_chn;
 772         int ret = 0;
 773         bool recovering = false;
 774
 775         if (!nvdev || nvdev->destroy)
 776                 return -ENODEV;
 777
 778         num_chn = nvdev->num_chn;
 779         max_chn = min_t(u32, nvdev->max_chn, num_online_cpus());
 780
 781         if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5) {
 782                 pr_info("vRSS unsupported before NVSP Version 5\n");
 783                 return -EINVAL;
 784         }
 785
 786         /* We do not support rx, tx, or other */
 787         if (!channels ||
 788             channels->rx_count ||
 789             channels->tx_count ||
 790             channels->other_count ||
 791             (channels->combined_count < 1))
 792                 return -EINVAL;
 793
 794         if (channels->combined_count > max_chn) {
 795                 pr_info("combined channels too high, using %d\n", max_chn);
 796                 channels->combined_count = max_chn;
 797         }
 798
 799         ret = netvsc_close(net);
 800         if (ret)
 801                 goto out;
 802
 803  do_set:
 804         nvdev->start_remove = true;
 805         rndis_filter_device_remove(dev);
 806
 807         nvdev->num_chn = channels->combined_count;
 808
 809         net_device_ctx->device_ctx = dev;
 810         hv_set_drvdata(dev, net);
 811
 812         memset(&device_info, 0, sizeof(device_info));
 813         device_info.num_chn = nvdev->num_chn; /* passed to RNDIS */
 814         device_info.ring_size = ring_size;
 815         device_info.max_num_vrss_chns = max_num_vrss_chns;
 816
 817         ret = rndis_filter_device_add(dev, &device_info);
 818         if (ret) {
 819                 if (recovering) {
 820                         netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
 821                         return ret;
 822                 }
 823                 goto recover;
 824         }
 825
 826         nvdev = hv_get_drvdata(dev);
 827
 828         ret = netif_set_real_num_tx_queues(net, nvdev->num_chn);
 829         if (ret) {
 830                 if (recovering) {
 831                         netdev_err(net, "could not set tx queue count (ret %d)\n", ret);
 832                         return ret;
 833                 }
 834                 goto recover;
 835         }
 836
 837         ret = netif_set_real_num_rx_queues(net, nvdev->num_chn);
 838         if (ret) {
 839                 if (recovering) {
 840                         netdev_err(net, "could not set rx queue count (ret %d)\n", ret);
 841                         return ret;
 842                 }
 843                 goto recover;
 844         }
 845
 846  out:
 847         netvsc_open(net);
 848
 849         return ret;
 850
 851  recover:
 852         /* If the above failed, we attempt to recover through the same
 853          * process but with the original number of channels.
 854          */
 855         netdev_err(net, "could not set channels, recovering\n");
 856         recovering = true;
 857         channels->combined_count = num_chn;
 858         goto do_set;
 859 }
 860
 861 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
 862 {
 863         struct net_device_context *ndevctx = netdev_priv(ndev);
 864         struct hv_device *hdev =  ndevctx->device_ctx;
 865         struct netvsc_device *nvdev = hv_get_drvdata(hdev);
 866         struct netvsc_device_info device_info;
 867         int limit = ETH_DATA_LEN;
 868         int ret = 0;
 869
 870         if (nvdev == NULL || nvdev->destroy)
 871                 return -ENODEV;
 872
 873         if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
 874                 limit = NETVSC_MTU - ETH_HLEN;
 875
 876         if (mtu < NETVSC_MTU_MIN || mtu > limit)
 877                 return -EINVAL;
 878
 879         ret = netvsc_close(ndev);
 880         if (ret)
 881                 goto out;
 882
 883         nvdev->start_remove = true;
 884         rndis_filter_device_remove(hdev);
 885
 886         ndev->mtu = mtu;
 887
 888         ndevctx->device_ctx = hdev;
 889         hv_set_drvdata(hdev, ndev);
 890
 891         memset(&device_info, 0, sizeof(device_info));
 892         device_info.ring_size = ring_size;
 893         device_info.num_chn = nvdev->num_chn;
 894         device_info.max_num_vrss_chns = max_num_vrss_chns;
 895         rndis_filter_device_add(hdev, &device_info);
 896
 897 out:
 898         netvsc_open(ndev);
 899
 900         return ret;
 901 }
 902
 903 static struct rtnl_link_stats64 *netvsc_get_stats64(struct net_device *net,
 904                                                     struct rtnl_link_stats64 *t)
 905 {
 906         struct net_device_context *ndev_ctx = netdev_priv(net);
 907         int cpu;
 908
 909         for_each_possible_cpu(cpu) {
 910                 struct netvsc_stats *tx_stats = per_cpu_ptr(ndev_ctx->tx_stats,
 911                                                             cpu);
 912                 struct netvsc_stats *rx_stats = per_cpu_ptr(ndev_ctx->rx_stats,
 913                                                             cpu);
 914                 u64 tx_packets, tx_bytes, rx_packets, rx_bytes;
 915                 unsigned int start;
 916
 917                 do {
 918                         start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
 919                         tx_packets = tx_stats->packets;
 920                         tx_bytes = tx_stats->bytes;
 921                 } while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
 922
 923                 do {
 924                         start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
 925                         rx_packets = rx_stats->packets;
 926                         rx_bytes = rx_stats->bytes;
 927                 } while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
 928
 929                 t->tx_bytes     += tx_bytes;
 930                 t->tx_packets   += tx_packets;
 931                 t->rx_bytes     += rx_bytes;
 932                 t->rx_packets   += rx_packets;
 933         }
 934
 935         t->tx_dropped   = net->stats.tx_dropped;
 936         t->tx_errors    = net->stats.tx_dropped;
 937
 938         t->rx_dropped   = net->stats.rx_dropped;
 939         t->rx_errors    = net->stats.rx_errors;
 940
 941         return t;
 942 }
 943
 944 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
 945 {
 946         struct net_device_context *ndevctx = netdev_priv(ndev);
 947         struct hv_device *hdev =  ndevctx->device_ctx;
 948         struct sockaddr *addr = p;
 949         char save_adr[ETH_ALEN];
 950         unsigned char save_aatype;
 951         int err;
 952
 953         memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
 954         save_aatype = ndev->addr_assign_type;
 955
 956         err = eth_mac_addr(ndev, p);
 957         if (err != 0)
 958                 return err;
 959
 960         err = rndis_filter_set_device_mac(hdev, addr->sa_data);
 961         if (err != 0) {
 962                 /* roll back to saved MAC */
 963                 memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
 964                 ndev->addr_assign_type = save_aatype;
 965         }
 966
 967         return err;
 968 }
 969
 970 #ifdef CONFIG_NET_POLL_CONTROLLER
 971 static void netvsc_poll_controller(struct net_device *net)
 972 {
 973         /* As netvsc_start_xmit() works synchronous we don't have to
 974          * trigger anything here.
 975          */
 976 }
 977 #endif
 978
 979 static const struct ethtool_ops ethtool_ops = {
 980         .get_drvinfo    = netvsc_get_drvinfo,
 981         .get_link       = ethtool_op_get_link,
 982         .get_channels   = netvsc_get_channels,
 983         .set_channels   = netvsc_set_channels,
 984 };
 985
 986 static const struct net_device_ops device_ops = {
 987         .ndo_open =                     netvsc_open,
 988         .ndo_stop =                     netvsc_close,
 989         .ndo_start_xmit =               netvsc_start_xmit,
 990         .ndo_set_rx_mode =              netvsc_set_multicast_list,
 991         .ndo_change_mtu =               netvsc_change_mtu,
 992         .ndo_validate_addr =            eth_validate_addr,
 993         .ndo_set_mac_address =          netvsc_set_mac_addr,
 994         .ndo_select_queue =             netvsc_select_queue,
 995         .ndo_get_stats64 =              netvsc_get_stats64,
 996 #ifdef CONFIG_NET_POLL_CONTROLLER
 997         .ndo_poll_controller =          netvsc_poll_controller,
 998 #endif
 999 };
1000
1001 /*
1002  * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
1003  * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
1004  * present send GARP packet to network peers with netif_notify_peers().
1005  */
1006 static void netvsc_link_change(struct work_struct *w)
1007 {
1008         struct net_device_context *ndev_ctx;
1009         struct net_device *net;
1010         struct netvsc_device *net_device;
1011         struct rndis_device *rdev;
1012         struct netvsc_reconfig *event = NULL;
1013         bool notify = false, reschedule = false;
1014         unsigned long flags, next_reconfig, delay;
1015
1016         ndev_ctx = container_of(w, struct net_device_context, dwork.work);
1017         net_device = hv_get_drvdata(ndev_ctx->device_ctx);
1018         rdev = net_device->extension;
1019         net = net_device->ndev;
1020
1021         next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
1022         if (time_is_after_jiffies(next_reconfig)) {
1023                 /* link_watch only sends one notification with current state
1024                  * per second, avoid doing reconfig more frequently. Handle
1025                  * wrap around.
1026                  */
1027                 delay = next_reconfig - jiffies;
1028                 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
1029                 schedule_delayed_work(&ndev_ctx->dwork, delay);
1030                 return;
1031         }
1032         ndev_ctx->last_reconfig = jiffies;
1033
1034         spin_lock_irqsave(&ndev_ctx->lock, flags);
1035         if (!list_empty(&ndev_ctx->reconfig_events)) {
1036                 event = list_first_entry(&ndev_ctx->reconfig_events,
1037                                          struct netvsc_reconfig, list);
1038                 list_del(&event->list);
1039                 reschedule = !list_empty(&ndev_ctx->reconfig_events);
1040         }
1041         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1042
1043         if (!event)
1044                 return;
1045
1046         rtnl_lock();
1047
1048         switch (event->event) {
1049                 /* Only the following events are possible due to the check in
1050                  * netvsc_linkstatus_callback()
1051                  */
1052         case RNDIS_STATUS_MEDIA_CONNECT:
1053                 if (rdev->link_state) {
1054                         rdev->link_state = false;
1055                         netif_carrier_on(net);
1056                         netif_tx_wake_all_queues(net);
1057                 } else {
1058                         notify = true;
1059                 }
1060                 kfree(event);
1061                 break;
1062         case RNDIS_STATUS_MEDIA_DISCONNECT:
1063                 if (!rdev->link_state) {
1064                         rdev->link_state = true;
1065                         netif_carrier_off(net);
1066                         netif_tx_stop_all_queues(net);
1067                 }
1068                 kfree(event);
1069                 break;
1070         case RNDIS_STATUS_NETWORK_CHANGE:
1071                 /* Only makes sense if carrier is present */
1072                 if (!rdev->link_state) {
1073                         rdev->link_state = true;
1074                         netif_carrier_off(net);
1075                         netif_tx_stop_all_queues(net);
1076                         event->event = RNDIS_STATUS_MEDIA_CONNECT;
1077                         spin_lock_irqsave(&ndev_ctx->lock, flags);
1078                         list_add_tail(&event->list, &ndev_ctx->reconfig_events);
1079                         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1080                         reschedule = true;
1081                 }
1082                 break;
1083         }
1084
1085         rtnl_unlock();
1086
1087         if (notify)
1088                 netdev_notify_peers(net);
1089
1090         /* link_watch only sends one notification with current state per
1091          * second, handle next reconfig event in 2 seconds.
1092          */
1093         if (reschedule)
1094                 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1095 }
1096
1097 static void netvsc_free_netdev(struct net_device *netdev)
1098 {
1099         struct net_device_context *net_device_ctx = netdev_priv(netdev);
1100
1101         free_percpu(net_device_ctx->tx_stats);
1102         free_percpu(net_device_ctx->rx_stats);
1103         free_netdev(netdev);
1104 }
1105
1106 static int netvsc_probe(struct hv_device *dev,
1107                         const struct hv_vmbus_device_id *dev_id)
1108 {
1109         struct net_device *net = NULL;
1110         struct net_device_context *net_device_ctx;
1111         struct netvsc_device_info device_info;
1112         struct netvsc_device *nvdev;
1113         int ret;
1114
1115         net = alloc_etherdev_mq(sizeof(struct net_device_context),
1116                                 num_online_cpus());
1117         if (!net)
1118                 return -ENOMEM;
1119
1120         netif_carrier_off(net);
1121
1122         net_device_ctx = netdev_priv(net);
1123         net_device_ctx->device_ctx = dev;
1124         net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
1125         if (netif_msg_probe(net_device_ctx))
1126                 netdev_dbg(net, "netvsc msg_enable: %d\n",
1127                            net_device_ctx->msg_enable);
1128
1129         net_device_ctx->tx_stats = netdev_alloc_pcpu_stats(struct netvsc_stats);
1130         if (!net_device_ctx->tx_stats) {
1131                 free_netdev(net);
1132                 return -ENOMEM;
1133         }
1134         net_device_ctx->rx_stats = netdev_alloc_pcpu_stats(struct netvsc_stats);
1135         if (!net_device_ctx->rx_stats) {
1136                 free_percpu(net_device_ctx->tx_stats);
1137                 free_netdev(net);
1138                 return -ENOMEM;
1139         }
1140
1141         hv_set_drvdata(dev, net);
1142         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
1143         INIT_WORK(&net_device_ctx->work, do_set_multicast);
1144
1145         spin_lock_init(&net_device_ctx->lock);
1146         INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
1147
1148         net->netdev_ops = &device_ops;
1149
1150         net->hw_features = NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_IP_CSUM |
1151                                 NETIF_F_TSO;
1152         net->features = NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_SG | NETIF_F_RXCSUM |
1153                         NETIF_F_IP_CSUM | NETIF_F_TSO;
1154
1155         net->ethtool_ops = &ethtool_ops;
1156         SET_NETDEV_DEV(net, &dev->device);
1157
1158         /* Notify the netvsc driver of the new device */
1159         memset(&device_info, 0, sizeof(device_info));
1160         device_info.ring_size = ring_size;
1161         device_info.max_num_vrss_chns = max_num_vrss_chns;
1162         ret = rndis_filter_device_add(dev, &device_info);
1163         if (ret != 0) {
1164                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
1165                 netvsc_free_netdev(net);
1166                 hv_set_drvdata(dev, NULL);
1167                 return ret;
1168         }
1169         memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
1170
1171         nvdev = hv_get_drvdata(dev);
1172         netif_set_real_num_tx_queues(net, nvdev->num_chn);
1173         netif_set_real_num_rx_queues(net, nvdev->num_chn);
1174
1175         ret = register_netdev(net);
1176         if (ret != 0) {
1177                 pr_err("Unable to register netdev.\n");
1178                 rndis_filter_device_remove(dev);
1179                 netvsc_free_netdev(net);
1180         }
1181
1182         return ret;
1183 }
1184
1185 static int netvsc_remove(struct hv_device *dev)
1186 {
1187         struct net_device *net;
1188         struct net_device_context *ndev_ctx;
1189         struct netvsc_device *net_device;
1190
1191         net_device = hv_get_drvdata(dev);
1192         net = net_device->ndev;
1193
1194         if (net == NULL) {
1195                 dev_err(&dev->device, "No net device to remove\n");
1196                 return 0;
1197         }
1198
1199         net_device->start_remove = true;
1200
1201         ndev_ctx = netdev_priv(net);
1202         cancel_delayed_work_sync(&ndev_ctx->dwork);
1203         cancel_work_sync(&ndev_ctx->work);
1204
1205         /* Stop outbound asap */
1206         netif_tx_disable(net);
1207
1208         unregister_netdev(net);
1209
1210         /*
1211          * Call to the vsc driver to let it know that the device is being
1212          * removed
1213          */
1214         rndis_filter_device_remove(dev);
1215
1216         netvsc_free_netdev(net);
1217         return 0;
1218 }
1219
1220 static const struct hv_vmbus_device_id id_table[] = {
1221         /* Network guid */
1222         { HV_NIC_GUID, },
1223         { },
1224 };
1225
1226 MODULE_DEVICE_TABLE(vmbus, id_table);
1227
1228 /* The one and only one */
1229 static struct  hv_driver netvsc_drv = {
1230         .name = KBUILD_MODNAME,
1231         .id_table = id_table,
1232         .probe = netvsc_probe,
1233         .remove = netvsc_remove,
1234 };
1235
1236 static void __exit netvsc_drv_exit(void)
1237 {
1238         vmbus_driver_unregister(&netvsc_drv);
1239 }
1240
1241 static int __init netvsc_drv_init(void)
1242 {
1243         if (ring_size < RING_SIZE_MIN) {
1244                 ring_size = RING_SIZE_MIN;
1245                 pr_info("Increased ring_size to %d (min allowed)\n",
1246                         ring_size);
1247         }
1248         return vmbus_driver_register(&netvsc_drv);
1249 }
1250
1251 MODULE_LICENSE("GPL");
1252 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
1253
1254 module_init(netvsc_drv_init);
1255 module_exit(netvsc_drv_exit);