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
 475         packet->q_idx = skb_get_queue_mapping(skb);
 476
 477         packet->total_data_buflen = skb->len;
 478
 479         rndis_msg = (struct rndis_message *)skb->head;
 480
 481         memset(rndis_msg, 0, RNDIS_AND_PPI_SIZE);
 482
 483         isvlan = skb->vlan_tci & VLAN_TAG_PRESENT;
 484
 485         /* Add the rndis header */
 486         rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
 487         rndis_msg->msg_len = packet->total_data_buflen;
 488         rndis_pkt = &rndis_msg->msg.pkt;
 489         rndis_pkt->data_offset = sizeof(struct rndis_packet);
 490         rndis_pkt->data_len = packet->total_data_buflen;
 491         rndis_pkt->per_pkt_info_offset = sizeof(struct rndis_packet);
 492
 493         rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
 494
 495         hash = skb_get_hash_raw(skb);
 496         if (hash != 0 && net->real_num_tx_queues > 1) {
 497                 rndis_msg_size += NDIS_HASH_PPI_SIZE;
 498                 ppi = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
 499                                     NBL_HASH_VALUE);
 500                 *(u32 *)((void *)ppi + ppi->ppi_offset) = hash;
 501         }
 502
 503         if (isvlan) {
 504                 struct ndis_pkt_8021q_info *vlan;
 505
 506                 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
 507                 ppi = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
 508                                         IEEE_8021Q_INFO);
 509                 vlan = (struct ndis_pkt_8021q_info *)((void *)ppi +
 510                                                 ppi->ppi_offset);
 511                 vlan->vlanid = skb->vlan_tci & VLAN_VID_MASK;
 512                 vlan->pri = (skb->vlan_tci & VLAN_PRIO_MASK) >>
 513                                 VLAN_PRIO_SHIFT;
 514         }
 515
 516         net_trans_info = get_net_transport_info(skb, &hdr_offset);
 517         if (net_trans_info == TRANSPORT_INFO_NOT_IP)
 518                 goto do_send;
 519
 520         /*
 521          * Setup the sendside checksum offload only if this is not a
 522          * GSO packet.
 523          */
 524         if (skb_is_gso(skb))
 525                 goto do_lso;
 526
 527         if ((skb->ip_summed == CHECKSUM_NONE) ||
 528             (skb->ip_summed == CHECKSUM_UNNECESSARY))
 529                 goto do_send;
 530
 531         rndis_msg_size += NDIS_CSUM_PPI_SIZE;
 532         ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
 533                             TCPIP_CHKSUM_PKTINFO);
 534
 535         csum_info = (struct ndis_tcp_ip_checksum_info *)((void *)ppi +
 536                         ppi->ppi_offset);
 537
 538         if (net_trans_info & (INFO_IPV4 << 16))
 539                 csum_info->transmit.is_ipv4 = 1;
 540         else
 541                 csum_info->transmit.is_ipv6 = 1;
 542
 543         if (net_trans_info & INFO_TCP) {
 544                 csum_info->transmit.tcp_checksum = 1;
 545                 csum_info->transmit.tcp_header_offset = hdr_offset;
 546         } else if (net_trans_info & INFO_UDP) {
 547                 /* UDP checksum offload is not supported on ws2008r2.
 548                  * Furthermore, on ws2012 and ws2012r2, there are some
 549                  * issues with udp checksum offload from Linux guests.
 550                  * (these are host issues).
 551                  * For now compute the checksum here.
 552                  */
 553                 struct udphdr *uh;
 554                 u16 udp_len;
 555
 556                 ret = skb_cow_head(skb, 0);
 557                 if (ret)
 558                         goto drop;
 559
 560                 uh = udp_hdr(skb);
 561                 udp_len = ntohs(uh->len);
 562                 uh->check = 0;
 563                 uh->check = csum_tcpudp_magic(ip_hdr(skb)->saddr,
 564                                               ip_hdr(skb)->daddr,
 565                                               udp_len, IPPROTO_UDP,
 566                                               csum_partial(uh, udp_len, 0));
 567                 if (uh->check == 0)
 568                         uh->check = CSUM_MANGLED_0;
 569
 570                 csum_info->transmit.udp_checksum = 0;
 571         }
 572         goto do_send;
 573
 574 do_lso:
 575         rndis_msg_size += NDIS_LSO_PPI_SIZE;
 576         ppi = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
 577                             TCP_LARGESEND_PKTINFO);
 578
 579         lso_info = (struct ndis_tcp_lso_info *)((void *)ppi +
 580                         ppi->ppi_offset);
 581
 582         lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
 583         if (net_trans_info & (INFO_IPV4 << 16)) {
 584                 lso_info->lso_v2_transmit.ip_version =
 585                         NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
 586                 ip_hdr(skb)->tot_len = 0;
 587                 ip_hdr(skb)->check = 0;
 588                 tcp_hdr(skb)->check =
 589                 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
 590                                    ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 591         } else {
 592                 lso_info->lso_v2_transmit.ip_version =
 593                         NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
 594                 ipv6_hdr(skb)->payload_len = 0;
 595                 tcp_hdr(skb)->check =
 596                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 597                                 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 598         }
 599         lso_info->lso_v2_transmit.tcp_header_offset = hdr_offset;
 600         lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
 601
 602 do_send:
 603         /* Start filling in the page buffers with the rndis hdr */
 604         rndis_msg->msg_len += rndis_msg_size;
 605         packet->total_data_buflen = rndis_msg->msg_len;
 606         packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
 607                                                skb, packet, &pb);
 608
 609         ret = netvsc_send(net_device_ctx->device_ctx, packet,
 610                           rndis_msg, &pb, skb);
 611
 612 drop:
 613         if (ret == 0) {
 614                 u64_stats_update_begin(&tx_stats->syncp);
 615                 tx_stats->packets++;
 616                 tx_stats->bytes += skb_length;
 617                 u64_stats_update_end(&tx_stats->syncp);
 618         } else {
 619                 if (ret != -EAGAIN) {
 620                         dev_kfree_skb_any(skb);
 621                         net->stats.tx_dropped++;
 622                 }
 623         }
 624
 625         return (ret == -EAGAIN) ? NETDEV_TX_BUSY : NETDEV_TX_OK;
 626 }
 627
 628 /*
 629  * netvsc_linkstatus_callback - Link up/down notification
 630  */
 631 void netvsc_linkstatus_callback(struct hv_device *device_obj,
 632                                 struct rndis_message *resp)
 633 {
 634         struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
 635         struct net_device *net;
 636         struct net_device_context *ndev_ctx;
 637         struct netvsc_device *net_device;
 638         struct netvsc_reconfig *event;
 639         unsigned long flags;
 640
 641         /* Handle link change statuses only */
 642         if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
 643             indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
 644             indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
 645                 return;
 646
 647         net_device = hv_get_drvdata(device_obj);
 648         net = net_device->ndev;
 649
 650         if (!net || net->reg_state != NETREG_REGISTERED)
 651                 return;
 652
 653         ndev_ctx = netdev_priv(net);
 654
 655         event = kzalloc(sizeof(*event), GFP_ATOMIC);
 656         if (!event)
 657                 return;
 658         event->event = indicate->status;
 659
 660         spin_lock_irqsave(&ndev_ctx->lock, flags);
 661         list_add_tail(&event->list, &ndev_ctx->reconfig_events);
 662         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
 663
 664         schedule_delayed_work(&ndev_ctx->dwork, 0);
 665 }
 666
 667 /*
 668  * netvsc_recv_callback -  Callback when we receive a packet from the
 669  * "wire" on the specified device.
 670  */
 671 int netvsc_recv_callback(struct hv_device *device_obj,
 672                                 struct hv_netvsc_packet *packet,
 673                                 void **data,
 674                                 struct ndis_tcp_ip_checksum_info *csum_info,
 675                                 struct vmbus_channel *channel,
 676                                 u16 vlan_tci)
 677 {
 678         struct net_device *net;
 679         struct net_device_context *net_device_ctx;
 680         struct sk_buff *skb;
 681         struct netvsc_stats *rx_stats;
 682
 683         net = ((struct netvsc_device *)hv_get_drvdata(device_obj))->ndev;
 684         if (!net || net->reg_state != NETREG_REGISTERED) {
 685                 return NVSP_STAT_FAIL;
 686         }
 687         net_device_ctx = netdev_priv(net);
 688         rx_stats = this_cpu_ptr(net_device_ctx->rx_stats);
 689
 690         /* Allocate a skb - TODO direct I/O to pages? */
 691         skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
 692         if (unlikely(!skb)) {
 693                 ++net->stats.rx_dropped;
 694                 return NVSP_STAT_FAIL;
 695         }
 696
 697         /*
 698          * Copy to skb. This copy is needed here since the memory pointed by
 699          * hv_netvsc_packet cannot be deallocated
 700          */
 701         memcpy(skb_put(skb, packet->total_data_buflen), *data,
 702                 packet->total_data_buflen);
 703
 704         skb->protocol = eth_type_trans(skb, net);
 705         if (csum_info) {
 706                 /* We only look at the IP checksum here.
 707                  * Should we be dropping the packet if checksum
 708                  * failed? How do we deal with other checksums - TCP/UDP?
 709                  */
 710                 if (csum_info->receive.ip_checksum_succeeded)
 711                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 712                 else
 713                         skb->ip_summed = CHECKSUM_NONE;
 714         }
 715
 716         if (vlan_tci & VLAN_TAG_PRESENT)
 717                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 718                                        vlan_tci);
 719
 720         skb_record_rx_queue(skb, channel->
 721                             offermsg.offer.sub_channel_index);
 722
 723         u64_stats_update_begin(&rx_stats->syncp);
 724         rx_stats->packets++;
 725         rx_stats->bytes += packet->total_data_buflen;
 726         u64_stats_update_end(&rx_stats->syncp);
 727
 728         /*
 729          * Pass the skb back up. Network stack will deallocate the skb when it
 730          * is done.
 731          * TODO - use NAPI?
 732          */
 733         netif_rx(skb);
 734
 735         return 0;
 736 }
 737
 738 static void netvsc_get_drvinfo(struct net_device *net,
 739                                struct ethtool_drvinfo *info)
 740 {
 741         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
 742         strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
 743 }
 744
 745 static void netvsc_get_channels(struct net_device *net,
 746                                 struct ethtool_channels *channel)
 747 {
 748         struct net_device_context *net_device_ctx = netdev_priv(net);
 749         struct hv_device *dev = net_device_ctx->device_ctx;
 750         struct netvsc_device *nvdev = hv_get_drvdata(dev);
 751
 752         if (nvdev) {
 753                 channel->max_combined   = nvdev->max_chn;
 754                 channel->combined_count = nvdev->num_chn;
 755         }
 756 }
 757
 758 static int netvsc_set_channels(struct net_device *net,
 759                                struct ethtool_channels *channels)
 760 {
 761         struct net_device_context *net_device_ctx = netdev_priv(net);
 762         struct hv_device *dev = net_device_ctx->device_ctx;
 763         struct netvsc_device *nvdev = hv_get_drvdata(dev);
 764         struct netvsc_device_info device_info;
 765         u32 num_chn;
 766         u32 max_chn;
 767         int ret = 0;
 768         bool recovering = false;
 769
 770         if (!nvdev || nvdev->destroy)
 771                 return -ENODEV;
 772
 773         num_chn = nvdev->num_chn;
 774         max_chn = min_t(u32, nvdev->max_chn, num_online_cpus());
 775
 776         if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5) {
 777                 pr_info("vRSS unsupported before NVSP Version 5\n");
 778                 return -EINVAL;
 779         }
 780
 781         /* We do not support rx, tx, or other */
 782         if (!channels ||
 783             channels->rx_count ||
 784             channels->tx_count ||
 785             channels->other_count ||
 786             (channels->combined_count < 1))
 787                 return -EINVAL;
 788
 789         if (channels->combined_count > max_chn) {
 790                 pr_info("combined channels too high, using %d\n", max_chn);
 791                 channels->combined_count = max_chn;
 792         }
 793
 794         ret = netvsc_close(net);
 795         if (ret)
 796                 goto out;
 797
 798  do_set:
 799         nvdev->start_remove = true;
 800         rndis_filter_device_remove(dev);
 801
 802         nvdev->num_chn = channels->combined_count;
 803
 804         net_device_ctx->device_ctx = dev;
 805         hv_set_drvdata(dev, net);
 806
 807         memset(&device_info, 0, sizeof(device_info));
 808         device_info.num_chn = nvdev->num_chn; /* passed to RNDIS */
 809         device_info.ring_size = ring_size;
 810         device_info.max_num_vrss_chns = max_num_vrss_chns;
 811
 812         ret = rndis_filter_device_add(dev, &device_info);
 813         if (ret) {
 814                 if (recovering) {
 815                         netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
 816                         return ret;
 817                 }
 818                 goto recover;
 819         }
 820
 821         nvdev = hv_get_drvdata(dev);
 822
 823         ret = netif_set_real_num_tx_queues(net, nvdev->num_chn);
 824         if (ret) {
 825                 if (recovering) {
 826                         netdev_err(net, "could not set tx queue count (ret %d)\n", ret);
 827                         return ret;
 828                 }
 829                 goto recover;
 830         }
 831
 832         ret = netif_set_real_num_rx_queues(net, nvdev->num_chn);
 833         if (ret) {
 834                 if (recovering) {
 835                         netdev_err(net, "could not set rx queue count (ret %d)\n", ret);
 836                         return ret;
 837                 }
 838                 goto recover;
 839         }
 840
 841  out:
 842         netvsc_open(net);
 843
 844         return ret;
 845
 846  recover:
 847         /* If the above failed, we attempt to recover through the same
 848          * process but with the original number of channels.
 849          */
 850         netdev_err(net, "could not set channels, recovering\n");
 851         recovering = true;
 852         channels->combined_count = num_chn;
 853         goto do_set;
 854 }
 855
 856 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
 857 {
 858         struct net_device_context *ndevctx = netdev_priv(ndev);
 859         struct hv_device *hdev =  ndevctx->device_ctx;
 860         struct netvsc_device *nvdev = hv_get_drvdata(hdev);
 861         struct netvsc_device_info device_info;
 862         int limit = ETH_DATA_LEN;
 863         int ret = 0;
 864
 865         if (nvdev == NULL || nvdev->destroy)
 866                 return -ENODEV;
 867
 868         if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
 869                 limit = NETVSC_MTU - ETH_HLEN;
 870
 871         if (mtu < NETVSC_MTU_MIN || mtu > limit)
 872                 return -EINVAL;
 873
 874         ret = netvsc_close(ndev);
 875         if (ret)
 876                 goto out;
 877
 878         nvdev->start_remove = true;
 879         rndis_filter_device_remove(hdev);
 880
 881         ndev->mtu = mtu;
 882
 883         ndevctx->device_ctx = hdev;
 884         hv_set_drvdata(hdev, ndev);
 885
 886         memset(&device_info, 0, sizeof(device_info));
 887         device_info.ring_size = ring_size;
 888         device_info.num_chn = nvdev->num_chn;
 889         device_info.max_num_vrss_chns = max_num_vrss_chns;
 890         rndis_filter_device_add(hdev, &device_info);
 891
 892 out:
 893         netvsc_open(ndev);
 894
 895         return ret;
 896 }
 897
 898 static struct rtnl_link_stats64 *netvsc_get_stats64(struct net_device *net,
 899                                                     struct rtnl_link_stats64 *t)
 900 {
 901         struct net_device_context *ndev_ctx = netdev_priv(net);
 902         int cpu;
 903
 904         for_each_possible_cpu(cpu) {
 905                 struct netvsc_stats *tx_stats = per_cpu_ptr(ndev_ctx->tx_stats,
 906                                                             cpu);
 907                 struct netvsc_stats *rx_stats = per_cpu_ptr(ndev_ctx->rx_stats,
 908                                                             cpu);
 909                 u64 tx_packets, tx_bytes, rx_packets, rx_bytes;
 910                 unsigned int start;
 911
 912                 do {
 913                         start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
 914                         tx_packets = tx_stats->packets;
 915                         tx_bytes = tx_stats->bytes;
 916                 } while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
 917
 918                 do {
 919                         start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
 920                         rx_packets = rx_stats->packets;
 921                         rx_bytes = rx_stats->bytes;
 922                 } while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
 923
 924                 t->tx_bytes     += tx_bytes;
 925                 t->tx_packets   += tx_packets;
 926                 t->rx_bytes     += rx_bytes;
 927                 t->rx_packets   += rx_packets;
 928         }
 929
 930         t->tx_dropped   = net->stats.tx_dropped;
 931         t->tx_errors    = net->stats.tx_dropped;
 932
 933         t->rx_dropped   = net->stats.rx_dropped;
 934         t->rx_errors    = net->stats.rx_errors;
 935
 936         return t;
 937 }
 938
 939 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
 940 {
 941         struct net_device_context *ndevctx = netdev_priv(ndev);
 942         struct hv_device *hdev =  ndevctx->device_ctx;
 943         struct sockaddr *addr = p;
 944         char save_adr[ETH_ALEN];
 945         unsigned char save_aatype;
 946         int err;
 947
 948         memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
 949         save_aatype = ndev->addr_assign_type;
 950
 951         err = eth_mac_addr(ndev, p);
 952         if (err != 0)
 953                 return err;
 954
 955         err = rndis_filter_set_device_mac(hdev, addr->sa_data);
 956         if (err != 0) {
 957                 /* roll back to saved MAC */
 958                 memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
 959                 ndev->addr_assign_type = save_aatype;
 960         }
 961
 962         return err;
 963 }
 964
 965 #ifdef CONFIG_NET_POLL_CONTROLLER
 966 static void netvsc_poll_controller(struct net_device *net)
 967 {
 968         /* As netvsc_start_xmit() works synchronous we don't have to
 969          * trigger anything here.
 970          */
 971 }
 972 #endif
 973
 974 static const struct ethtool_ops ethtool_ops = {
 975         .get_drvinfo    = netvsc_get_drvinfo,
 976         .get_link       = ethtool_op_get_link,
 977         .get_channels   = netvsc_get_channels,
 978         .set_channels   = netvsc_set_channels,
 979 };
 980
 981 static const struct net_device_ops device_ops = {
 982         .ndo_open =                     netvsc_open,
 983         .ndo_stop =                     netvsc_close,
 984         .ndo_start_xmit =               netvsc_start_xmit,
 985         .ndo_set_rx_mode =              netvsc_set_multicast_list,
 986         .ndo_change_mtu =               netvsc_change_mtu,
 987         .ndo_validate_addr =            eth_validate_addr,
 988         .ndo_set_mac_address =          netvsc_set_mac_addr,
 989         .ndo_select_queue =             netvsc_select_queue,
 990         .ndo_get_stats64 =              netvsc_get_stats64,
 991 #ifdef CONFIG_NET_POLL_CONTROLLER
 992         .ndo_poll_controller =          netvsc_poll_controller,
 993 #endif
 994 };
 995
 996 /*
 997  * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
 998  * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
 999  * present send GARP packet to network peers with netif_notify_peers().
1000  */
1001 static void netvsc_link_change(struct work_struct *w)
1002 {
1003         struct net_device_context *ndev_ctx;
1004         struct net_device *net;
1005         struct netvsc_device *net_device;
1006         struct rndis_device *rdev;
1007         struct netvsc_reconfig *event = NULL;
1008         bool notify = false, reschedule = false;
1009         unsigned long flags, next_reconfig, delay;
1010
1011         ndev_ctx = container_of(w, struct net_device_context, dwork.work);
1012         net_device = hv_get_drvdata(ndev_ctx->device_ctx);
1013         rdev = net_device->extension;
1014         net = net_device->ndev;
1015
1016         next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
1017         if (time_is_after_jiffies(next_reconfig)) {
1018                 /* link_watch only sends one notification with current state
1019                  * per second, avoid doing reconfig more frequently. Handle
1020                  * wrap around.
1021                  */
1022                 delay = next_reconfig - jiffies;
1023                 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
1024                 schedule_delayed_work(&ndev_ctx->dwork, delay);
1025                 return;
1026         }
1027         ndev_ctx->last_reconfig = jiffies;
1028
1029         spin_lock_irqsave(&ndev_ctx->lock, flags);
1030         if (!list_empty(&ndev_ctx->reconfig_events)) {
1031                 event = list_first_entry(&ndev_ctx->reconfig_events,
1032                                          struct netvsc_reconfig, list);
1033                 list_del(&event->list);
1034                 reschedule = !list_empty(&ndev_ctx->reconfig_events);
1035         }
1036         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1037
1038         if (!event)
1039                 return;
1040
1041         rtnl_lock();
1042
1043         switch (event->event) {
1044                 /* Only the following events are possible due to the check in
1045                  * netvsc_linkstatus_callback()
1046                  */
1047         case RNDIS_STATUS_MEDIA_CONNECT:
1048                 if (rdev->link_state) {
1049                         rdev->link_state = false;
1050                         netif_carrier_on(net);
1051                         netif_tx_wake_all_queues(net);
1052                 } else {
1053                         notify = true;
1054                 }
1055                 kfree(event);
1056                 break;
1057         case RNDIS_STATUS_MEDIA_DISCONNECT:
1058                 if (!rdev->link_state) {
1059                         rdev->link_state = true;
1060                         netif_carrier_off(net);
1061                         netif_tx_stop_all_queues(net);
1062                 }
1063                 kfree(event);
1064                 break;
1065         case RNDIS_STATUS_NETWORK_CHANGE:
1066                 /* Only makes sense if carrier is present */
1067                 if (!rdev->link_state) {
1068                         rdev->link_state = true;
1069                         netif_carrier_off(net);
1070                         netif_tx_stop_all_queues(net);
1071                         event->event = RNDIS_STATUS_MEDIA_CONNECT;
1072                         spin_lock_irqsave(&ndev_ctx->lock, flags);
1073                         list_add_tail(&event->list, &ndev_ctx->reconfig_events);
1074                         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1075                         reschedule = true;
1076                 }
1077                 break;
1078         }
1079
1080         rtnl_unlock();
1081
1082         if (notify)
1083                 netdev_notify_peers(net);
1084
1085         /* link_watch only sends one notification with current state per
1086          * second, handle next reconfig event in 2 seconds.
1087          */
1088         if (reschedule)
1089                 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1090 }
1091
1092 static void netvsc_free_netdev(struct net_device *netdev)
1093 {
1094         struct net_device_context *net_device_ctx = netdev_priv(netdev);
1095
1096         free_percpu(net_device_ctx->tx_stats);
1097         free_percpu(net_device_ctx->rx_stats);
1098         free_netdev(netdev);
1099 }
1100
1101 static int netvsc_probe(struct hv_device *dev,
1102                         const struct hv_vmbus_device_id *dev_id)
1103 {
1104         struct net_device *net = NULL;
1105         struct net_device_context *net_device_ctx;
1106         struct netvsc_device_info device_info;
1107         struct netvsc_device *nvdev;
1108         int ret;
1109
1110         net = alloc_etherdev_mq(sizeof(struct net_device_context),
1111                                 num_online_cpus());
1112         if (!net)
1113                 return -ENOMEM;
1114
1115         netif_carrier_off(net);
1116
1117         net_device_ctx = netdev_priv(net);
1118         net_device_ctx->device_ctx = dev;
1119         net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
1120         if (netif_msg_probe(net_device_ctx))
1121                 netdev_dbg(net, "netvsc msg_enable: %d\n",
1122                            net_device_ctx->msg_enable);
1123
1124         net_device_ctx->tx_stats = netdev_alloc_pcpu_stats(struct netvsc_stats);
1125         if (!net_device_ctx->tx_stats) {
1126                 free_netdev(net);
1127                 return -ENOMEM;
1128         }
1129         net_device_ctx->rx_stats = netdev_alloc_pcpu_stats(struct netvsc_stats);
1130         if (!net_device_ctx->rx_stats) {
1131                 free_percpu(net_device_ctx->tx_stats);
1132                 free_netdev(net);
1133                 return -ENOMEM;
1134         }
1135
1136         hv_set_drvdata(dev, net);
1137         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
1138         INIT_WORK(&net_device_ctx->work, do_set_multicast);
1139
1140         spin_lock_init(&net_device_ctx->lock);
1141         INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
1142
1143         net->netdev_ops = &device_ops;
1144
1145         net->hw_features = NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_IP_CSUM |
1146                                 NETIF_F_TSO;
1147         net->features = NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_SG | NETIF_F_RXCSUM |
1148                         NETIF_F_IP_CSUM | NETIF_F_TSO;
1149
1150         net->ethtool_ops = &ethtool_ops;
1151         SET_NETDEV_DEV(net, &dev->device);
1152
1153         /* Notify the netvsc driver of the new device */
1154         memset(&device_info, 0, sizeof(device_info));
1155         device_info.ring_size = ring_size;
1156         device_info.max_num_vrss_chns = max_num_vrss_chns;
1157         ret = rndis_filter_device_add(dev, &device_info);
1158         if (ret != 0) {
1159                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
1160                 netvsc_free_netdev(net);
1161                 hv_set_drvdata(dev, NULL);
1162                 return ret;
1163         }
1164         memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
1165
1166         nvdev = hv_get_drvdata(dev);
1167         netif_set_real_num_tx_queues(net, nvdev->num_chn);
1168         netif_set_real_num_rx_queues(net, nvdev->num_chn);
1169
1170         ret = register_netdev(net);
1171         if (ret != 0) {
1172                 pr_err("Unable to register netdev.\n");
1173                 rndis_filter_device_remove(dev);
1174                 netvsc_free_netdev(net);
1175         }
1176
1177         return ret;
1178 }
1179
1180 static int netvsc_remove(struct hv_device *dev)
1181 {
1182         struct net_device *net;
1183         struct net_device_context *ndev_ctx;
1184         struct netvsc_device *net_device;
1185
1186         net_device = hv_get_drvdata(dev);
1187         net = net_device->ndev;
1188
1189         if (net == NULL) {
1190                 dev_err(&dev->device, "No net device to remove\n");
1191                 return 0;
1192         }
1193
1194         net_device->start_remove = true;
1195
1196         ndev_ctx = netdev_priv(net);
1197         cancel_delayed_work_sync(&ndev_ctx->dwork);
1198         cancel_work_sync(&ndev_ctx->work);
1199
1200         /* Stop outbound asap */
1201         netif_tx_disable(net);
1202
1203         unregister_netdev(net);
1204
1205         /*
1206          * Call to the vsc driver to let it know that the device is being
1207          * removed
1208          */
1209         rndis_filter_device_remove(dev);
1210
1211         netvsc_free_netdev(net);
1212         return 0;
1213 }
1214
1215 static const struct hv_vmbus_device_id id_table[] = {
1216         /* Network guid */
1217         { HV_NIC_GUID, },
1218         { },
1219 };
1220
1221 MODULE_DEVICE_TABLE(vmbus, id_table);
1222
1223 /* The one and only one */
1224 static struct  hv_driver netvsc_drv = {
1225         .name = KBUILD_MODNAME,
1226         .id_table = id_table,
1227         .probe = netvsc_probe,
1228         .remove = netvsc_remove,
1229 };
1230
1231 static void __exit netvsc_drv_exit(void)
1232 {
1233         vmbus_driver_unregister(&netvsc_drv);
1234 }
1235
1236 static int __init netvsc_drv_init(void)
1237 {
1238         if (ring_size < RING_SIZE_MIN) {
1239                 ring_size = RING_SIZE_MIN;
1240                 pr_info("Increased ring_size to %d (min allowed)\n",
1241                         ring_size);
1242         }
1243         return vmbus_driver_register(&netvsc_drv);
1244 }
1245
1246 MODULE_LICENSE("GPL");
1247 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
1248
1249 module_init(netvsc_drv_init);
1250 module_exit(netvsc_drv_exit);