drivers/net/hyperv/rndis_filter.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (c) 2009, Microsoft Corporation.
   4  *
   5  * Authors:
   6  *   Haiyang Zhang <haiyangz@microsoft.com>
   7  *   Hank Janssen  <hjanssen@microsoft.com>
   8  */
   9 #include <linux/kernel.h>
  10 #include <linux/sched.h>
  11 #include <linux/wait.h>
  12 #include <linux/highmem.h>
  13 #include <linux/slab.h>
  14 #include <linux/io.h>
  15 #include <linux/if_ether.h>
  16 #include <linux/netdevice.h>
  17 #include <linux/if_vlan.h>
  18 #include <linux/nls.h>
  19 #include <linux/vmalloc.h>
  20 #include <linux/rtnetlink.h>
  21 #include <linux/ucs2_string.h>
  22
  23 #include "hyperv_net.h"
  24 #include "netvsc_trace.h"
  25
  26 static void rndis_set_multicast(struct work_struct *w);
  27
  28 #define RNDIS_EXT_LEN HV_HYP_PAGE_SIZE
  29 struct rndis_request {
  30         struct list_head list_ent;
  31         struct completion  wait_event;
  32
  33         struct rndis_message response_msg;
  34         /*
  35          * The buffer for extended info after the RNDIS response message. It's
  36          * referenced based on the data offset in the RNDIS message. Its size
  37          * is enough for current needs, and should be sufficient for the near
  38          * future.
  39          */
  40         u8 response_ext[RNDIS_EXT_LEN];
  41
  42         /* Simplify allocation by having a netvsc packet inline */
  43         struct hv_netvsc_packet pkt;
  44
  45         struct rndis_message request_msg;
  46         /*
  47          * The buffer for the extended info after the RNDIS request message.
  48          * It is referenced and sized in a similar way as response_ext.
  49          */
  50         u8 request_ext[RNDIS_EXT_LEN];
  51 };
  52
  53 static const u8 netvsc_hash_key[NETVSC_HASH_KEYLEN] = {
  54         0x6d, 0x5a, 0x56, 0xda, 0x25, 0x5b, 0x0e, 0xc2,
  55         0x41, 0x67, 0x25, 0x3d, 0x43, 0xa3, 0x8f, 0xb0,
  56         0xd0, 0xca, 0x2b, 0xcb, 0xae, 0x7b, 0x30, 0xb4,
  57         0x77, 0xcb, 0x2d, 0xa3, 0x80, 0x30, 0xf2, 0x0c,
  58         0x6a, 0x42, 0xb7, 0x3b, 0xbe, 0xac, 0x01, 0xfa
  59 };
  60
  61 static struct rndis_device *get_rndis_device(void)
  62 {
  63         struct rndis_device *device;
  64
  65         device = kzalloc(sizeof(struct rndis_device), GFP_KERNEL);
  66         if (!device)
  67                 return NULL;
  68
  69         spin_lock_init(&device->request_lock);
  70
  71         INIT_LIST_HEAD(&device->req_list);
  72         INIT_WORK(&device->mcast_work, rndis_set_multicast);
  73
  74         device->state = RNDIS_DEV_UNINITIALIZED;
  75
  76         return device;
  77 }
  78
  79 static struct rndis_request *get_rndis_request(struct rndis_device *dev,
  80                                              u32 msg_type,
  81                                              u32 msg_len)
  82 {
  83         struct rndis_request *request;
  84         struct rndis_message *rndis_msg;
  85         struct rndis_set_request *set;
  86         unsigned long flags;
  87
  88         request = kzalloc(sizeof(struct rndis_request), GFP_KERNEL);
  89         if (!request)
  90                 return NULL;
  91
  92         init_completion(&request->wait_event);
  93
  94         rndis_msg = &request->request_msg;
  95         rndis_msg->ndis_msg_type = msg_type;
  96         rndis_msg->msg_len = msg_len;
  97
  98         request->pkt.q_idx = 0;
  99
 100         /*
 101          * Set the request id. This field is always after the rndis header for
 102          * request/response packet types so we just used the SetRequest as a
 103          * template
 104          */
 105         set = &rndis_msg->msg.set_req;
 106         set->req_id = atomic_inc_return(&dev->new_req_id);
 107
 108         /* Add to the request list */
 109         spin_lock_irqsave(&dev->request_lock, flags);
 110         list_add_tail(&request->list_ent, &dev->req_list);
 111         spin_unlock_irqrestore(&dev->request_lock, flags);
 112
 113         return request;
 114 }
 115
 116 static void put_rndis_request(struct rndis_device *dev,
 117                             struct rndis_request *req)
 118 {
 119         unsigned long flags;
 120
 121         spin_lock_irqsave(&dev->request_lock, flags);
 122         list_del(&req->list_ent);
 123         spin_unlock_irqrestore(&dev->request_lock, flags);
 124
 125         kfree(req);
 126 }
 127
 128 static void dump_rndis_message(struct net_device *netdev,
 129                                const struct rndis_message *rndis_msg)
 130 {
 131         switch (rndis_msg->ndis_msg_type) {
 132         case RNDIS_MSG_PACKET:
 133                 netdev_dbg(netdev, "RNDIS_MSG_PACKET (len %u, "
 134                            "data offset %u data len %u, # oob %u, "
 135                            "oob offset %u, oob len %u, pkt offset %u, "
 136                            "pkt len %u\n",
 137                            rndis_msg->msg_len,
 138                            rndis_msg->msg.pkt.data_offset,
 139                            rndis_msg->msg.pkt.data_len,
 140                            rndis_msg->msg.pkt.num_oob_data_elements,
 141                            rndis_msg->msg.pkt.oob_data_offset,
 142                            rndis_msg->msg.pkt.oob_data_len,
 143                            rndis_msg->msg.pkt.per_pkt_info_offset,
 144                            rndis_msg->msg.pkt.per_pkt_info_len);
 145                 break;
 146
 147         case RNDIS_MSG_INIT_C:
 148                 netdev_dbg(netdev, "RNDIS_MSG_INIT_C "
 149                         "(len %u, id 0x%x, status 0x%x, major %d, minor %d, "
 150                         "device flags %d, max xfer size 0x%x, max pkts %u, "
 151                         "pkt aligned %u)\n",
 152                         rndis_msg->msg_len,
 153                         rndis_msg->msg.init_complete.req_id,
 154                         rndis_msg->msg.init_complete.status,
 155                         rndis_msg->msg.init_complete.major_ver,
 156                         rndis_msg->msg.init_complete.minor_ver,
 157                         rndis_msg->msg.init_complete.dev_flags,
 158                         rndis_msg->msg.init_complete.max_xfer_size,
 159                         rndis_msg->msg.init_complete.
 160                            max_pkt_per_msg,
 161                         rndis_msg->msg.init_complete.
 162                            pkt_alignment_factor);
 163                 break;
 164
 165         case RNDIS_MSG_QUERY_C:
 166                 netdev_dbg(netdev, "RNDIS_MSG_QUERY_C "
 167                         "(len %u, id 0x%x, status 0x%x, buf len %u, "
 168                         "buf offset %u)\n",
 169                         rndis_msg->msg_len,
 170                         rndis_msg->msg.query_complete.req_id,
 171                         rndis_msg->msg.query_complete.status,
 172                         rndis_msg->msg.query_complete.
 173                            info_buflen,
 174                         rndis_msg->msg.query_complete.
 175                            info_buf_offset);
 176                 break;
 177
 178         case RNDIS_MSG_SET_C:
 179                 netdev_dbg(netdev,
 180                         "RNDIS_MSG_SET_C (len %u, id 0x%x, status 0x%x)\n",
 181                         rndis_msg->msg_len,
 182                         rndis_msg->msg.set_complete.req_id,
 183                         rndis_msg->msg.set_complete.status);
 184                 break;
 185
 186         case RNDIS_MSG_INDICATE:
 187                 netdev_dbg(netdev, "RNDIS_MSG_INDICATE "
 188                         "(len %u, status 0x%x, buf len %u, buf offset %u)\n",
 189                         rndis_msg->msg_len,
 190                         rndis_msg->msg.indicate_status.status,
 191                         rndis_msg->msg.indicate_status.status_buflen,
 192                         rndis_msg->msg.indicate_status.status_buf_offset);
 193                 break;
 194
 195         default:
 196                 netdev_dbg(netdev, "0x%x (len %u)\n",
 197                         rndis_msg->ndis_msg_type,
 198                         rndis_msg->msg_len);
 199                 break;
 200         }
 201 }
 202
 203 static int rndis_filter_send_request(struct rndis_device *dev,
 204                                   struct rndis_request *req)
 205 {
 206         struct hv_netvsc_packet *packet;
 207         struct hv_page_buffer page_buf[2];
 208         struct hv_page_buffer *pb = page_buf;
 209         int ret;
 210
 211         /* Setup the packet to send it */
 212         packet = &req->pkt;
 213
 214         packet->total_data_buflen = req->request_msg.msg_len;
 215         packet->page_buf_cnt = 1;
 216
 217         pb[0].pfn = virt_to_phys(&req->request_msg) >>
 218                                         HV_HYP_PAGE_SHIFT;
 219         pb[0].len = req->request_msg.msg_len;
 220         pb[0].offset = offset_in_hvpage(&req->request_msg);
 221
 222         /* Add one page_buf when request_msg crossing page boundary */
 223         if (pb[0].offset + pb[0].len > HV_HYP_PAGE_SIZE) {
 224                 packet->page_buf_cnt++;
 225                 pb[0].len = HV_HYP_PAGE_SIZE -
 226                         pb[0].offset;
 227                 pb[1].pfn = virt_to_phys((void *)&req->request_msg
 228                         + pb[0].len) >> HV_HYP_PAGE_SHIFT;
 229                 pb[1].offset = 0;
 230                 pb[1].len = req->request_msg.msg_len -
 231                         pb[0].len;
 232         }
 233
 234         trace_rndis_send(dev->ndev, 0, &req->request_msg);
 235
 236         rcu_read_lock_bh();
 237         ret = netvsc_send(dev->ndev, packet, NULL, pb, NULL, false);
 238         rcu_read_unlock_bh();
 239
 240         return ret;
 241 }
 242
 243 static void rndis_set_link_state(struct rndis_device *rdev,
 244                                  struct rndis_request *request)
 245 {
 246         u32 link_status;
 247         struct rndis_query_complete *query_complete;
 248
 249         query_complete = &request->response_msg.msg.query_complete;
 250
 251         if (query_complete->status == RNDIS_STATUS_SUCCESS &&
 252             query_complete->info_buflen == sizeof(u32)) {
 253                 memcpy(&link_status, (void *)((unsigned long)query_complete +
 254                        query_complete->info_buf_offset), sizeof(u32));
 255                 rdev->link_state = link_status != 0;
 256         }
 257 }
 258
 259 static void rndis_filter_receive_response(struct net_device *ndev,
 260                                           struct netvsc_device *nvdev,
 261                                           const struct rndis_message *resp)
 262 {
 263         struct rndis_device *dev = nvdev->extension;
 264         struct rndis_request *request = NULL;
 265         bool found = false;
 266         unsigned long flags;
 267
 268         /* This should never happen, it means control message
 269          * response received after device removed.
 270          */
 271         if (dev->state == RNDIS_DEV_UNINITIALIZED) {
 272                 netdev_err(ndev,
 273                            "got rndis message uninitialized\n");
 274                 return;
 275         }
 276
 277         /* Ensure the packet is big enough to read req_id. Req_id is the 1st
 278          * field in any request/response message, so the payload should have at
 279          * least sizeof(u32) bytes
 280          */
 281         if (resp->msg_len - RNDIS_HEADER_SIZE < sizeof(u32)) {
 282                 netdev_err(ndev, "rndis msg_len too small: %u\n",
 283                            resp->msg_len);
 284                 return;
 285         }
 286
 287         spin_lock_irqsave(&dev->request_lock, flags);
 288         list_for_each_entry(request, &dev->req_list, list_ent) {
 289                 /*
 290                  * All request/response message contains RequestId as the 1st
 291                  * field
 292                  */
 293                 if (request->request_msg.msg.init_req.req_id
 294                     == resp->msg.init_complete.req_id) {
 295                         found = true;
 296                         break;
 297                 }
 298         }
 299         spin_unlock_irqrestore(&dev->request_lock, flags);
 300
 301         if (found) {
 302                 if (resp->msg_len <=
 303                     sizeof(struct rndis_message) + RNDIS_EXT_LEN) {
 304                         memcpy(&request->response_msg, resp,
 305                                resp->msg_len);
 306                         if (request->request_msg.ndis_msg_type ==
 307                             RNDIS_MSG_QUERY && request->request_msg.msg.
 308                             query_req.oid == RNDIS_OID_GEN_MEDIA_CONNECT_STATUS)
 309                                 rndis_set_link_state(dev, request);
 310                 } else {
 311                         netdev_err(ndev,
 312                                 "rndis response buffer overflow "
 313                                 "detected (size %u max %zu)\n",
 314                                 resp->msg_len,
 315                                 sizeof(struct rndis_message));
 316
 317                         if (resp->ndis_msg_type ==
 318                             RNDIS_MSG_RESET_C) {
 319                                 /* does not have a request id field */
 320                                 request->response_msg.msg.reset_complete.
 321                                         status = RNDIS_STATUS_BUFFER_OVERFLOW;
 322                         } else {
 323                                 request->response_msg.msg.
 324                                 init_complete.status =
 325                                         RNDIS_STATUS_BUFFER_OVERFLOW;
 326                         }
 327                 }
 328
 329                 complete(&request->wait_event);
 330         } else {
 331                 netdev_err(ndev,
 332                         "no rndis request found for this response "
 333                         "(id 0x%x res type 0x%x)\n",
 334                         resp->msg.init_complete.req_id,
 335                         resp->ndis_msg_type);
 336         }
 337 }
 338
 339 /*
 340  * Get the Per-Packet-Info with the specified type
 341  * return NULL if not found.
 342  */
 343 static inline void *rndis_get_ppi(struct net_device *ndev,
 344                                   struct rndis_packet *rpkt,
 345                                   u32 rpkt_len, u32 type, u8 internal)
 346 {
 347         struct rndis_per_packet_info *ppi;
 348         int len;
 349
 350         if (rpkt->per_pkt_info_offset == 0)
 351                 return NULL;
 352
 353         /* Validate info_offset and info_len */
 354         if (rpkt->per_pkt_info_offset < sizeof(struct rndis_packet) ||
 355             rpkt->per_pkt_info_offset > rpkt_len) {
 356                 netdev_err(ndev, "Invalid per_pkt_info_offset: %u\n",
 357                            rpkt->per_pkt_info_offset);
 358                 return NULL;
 359         }
 360
 361         if (rpkt->per_pkt_info_len > rpkt_len - rpkt->per_pkt_info_offset) {
 362                 netdev_err(ndev, "Invalid per_pkt_info_len: %u\n",
 363                            rpkt->per_pkt_info_len);
 364                 return NULL;
 365         }
 366
 367         ppi = (struct rndis_per_packet_info *)((ulong)rpkt +
 368                 rpkt->per_pkt_info_offset);
 369         len = rpkt->per_pkt_info_len;
 370
 371         while (len > 0) {
 372                 /* Validate ppi_offset and ppi_size */
 373                 if (ppi->size > len) {
 374                         netdev_err(ndev, "Invalid ppi size: %u\n", ppi->size);
 375                         continue;
 376                 }
 377
 378                 if (ppi->ppi_offset >= ppi->size) {
 379                         netdev_err(ndev, "Invalid ppi_offset: %u\n", ppi->ppi_offset);
 380                         continue;
 381                 }
 382
 383                 if (ppi->type == type && ppi->internal == internal)
 384                         return (void *)((ulong)ppi + ppi->ppi_offset);
 385                 len -= ppi->size;
 386                 ppi = (struct rndis_per_packet_info *)((ulong)ppi + ppi->size);
 387         }
 388
 389         return NULL;
 390 }
 391
 392 static inline
 393 void rsc_add_data(struct netvsc_channel *nvchan,
 394                   const struct ndis_pkt_8021q_info *vlan,
 395                   const struct ndis_tcp_ip_checksum_info *csum_info,
 396                   const u32 *hash_info,
 397                   void *data, u32 len)
 398 {
 399         u32 cnt = nvchan->rsc.cnt;
 400
 401         if (cnt) {
 402                 nvchan->rsc.pktlen += len;
 403         } else {
 404                 nvchan->rsc.vlan = vlan;
 405                 nvchan->rsc.csum_info = csum_info;
 406                 nvchan->rsc.pktlen = len;
 407                 nvchan->rsc.hash_info = hash_info;
 408         }
 409
 410         nvchan->rsc.data[cnt] = data;
 411         nvchan->rsc.len[cnt] = len;
 412         nvchan->rsc.cnt++;
 413 }
 414
 415 static int rndis_filter_receive_data(struct net_device *ndev,
 416                                      struct netvsc_device *nvdev,
 417                                      struct netvsc_channel *nvchan,
 418                                      struct rndis_message *msg,
 419                                      u32 data_buflen)
 420 {
 421         struct rndis_packet *rndis_pkt = &msg->msg.pkt;
 422         const struct ndis_tcp_ip_checksum_info *csum_info;
 423         const struct ndis_pkt_8021q_info *vlan;
 424         const struct rndis_pktinfo_id *pktinfo_id;
 425         const u32 *hash_info;
 426         u32 data_offset, rpkt_len;
 427         void *data;
 428         bool rsc_more = false;
 429         int ret;
 430
 431         /* Ensure data_buflen is big enough to read header fields */
 432         if (data_buflen < RNDIS_HEADER_SIZE + sizeof(struct rndis_packet)) {
 433                 netdev_err(ndev, "invalid rndis pkt, data_buflen too small: %u\n",
 434                            data_buflen);
 435                 return NVSP_STAT_FAIL;
 436         }
 437
 438         /* Validate rndis_pkt offset */
 439         if (rndis_pkt->data_offset >= data_buflen - RNDIS_HEADER_SIZE) {
 440                 netdev_err(ndev, "invalid rndis packet offset: %u\n",
 441                            rndis_pkt->data_offset);
 442                 return NVSP_STAT_FAIL;
 443         }
 444
 445         /* Remove the rndis header and pass it back up the stack */
 446         data_offset = RNDIS_HEADER_SIZE + rndis_pkt->data_offset;
 447
 448         rpkt_len = data_buflen - RNDIS_HEADER_SIZE;
 449         data_buflen -= data_offset;
 450
 451         /*
 452          * Make sure we got a valid RNDIS message, now total_data_buflen
 453          * should be the data packet size plus the trailer padding size
 454          */
 455         if (unlikely(data_buflen < rndis_pkt->data_len)) {
 456                 netdev_err(ndev, "rndis message buffer "
 457                            "overflow detected (got %u, min %u)"
 458                            "...dropping this message!\n",
 459                            data_buflen, rndis_pkt->data_len);
 460                 return NVSP_STAT_FAIL;
 461         }
 462
 463         vlan = rndis_get_ppi(ndev, rndis_pkt, rpkt_len, IEEE_8021Q_INFO, 0);
 464
 465         csum_info = rndis_get_ppi(ndev, rndis_pkt, rpkt_len, TCPIP_CHKSUM_PKTINFO, 0);
 466
 467         hash_info = rndis_get_ppi(ndev, rndis_pkt, rpkt_len, NBL_HASH_VALUE, 0);
 468
 469         pktinfo_id = rndis_get_ppi(ndev, rndis_pkt, rpkt_len, RNDIS_PKTINFO_ID, 1);
 470
 471         data = (void *)msg + data_offset;
 472
 473         /* Identify RSC frags, drop erroneous packets */
 474         if (pktinfo_id && (pktinfo_id->flag & RNDIS_PKTINFO_SUBALLOC)) {
 475                 if (pktinfo_id->flag & RNDIS_PKTINFO_1ST_FRAG)
 476                         nvchan->rsc.cnt = 0;
 477                 else if (nvchan->rsc.cnt == 0)
 478                         goto drop;
 479
 480                 rsc_more = true;
 481
 482                 if (pktinfo_id->flag & RNDIS_PKTINFO_LAST_FRAG)
 483                         rsc_more = false;
 484
 485                 if (rsc_more && nvchan->rsc.is_last)
 486                         goto drop;
 487         } else {
 488                 nvchan->rsc.cnt = 0;
 489         }
 490
 491         if (unlikely(nvchan->rsc.cnt >= NVSP_RSC_MAX))
 492                 goto drop;
 493
 494         /* Put data into per channel structure.
 495          * Also, remove the rndis trailer padding from rndis packet message
 496          * rndis_pkt->data_len tell us the real data length, we only copy
 497          * the data packet to the stack, without the rndis trailer padding
 498          */
 499         rsc_add_data(nvchan, vlan, csum_info, hash_info,
 500                      data, rndis_pkt->data_len);
 501
 502         if (rsc_more)
 503                 return NVSP_STAT_SUCCESS;
 504
 505         ret = netvsc_recv_callback(ndev, nvdev, nvchan);
 506         nvchan->rsc.cnt = 0;
 507
 508         return ret;
 509
 510 drop:
 511         /* Drop incomplete packet */
 512         nvchan->rsc.cnt = 0;
 513         return NVSP_STAT_FAIL;
 514 }
 515
 516 int rndis_filter_receive(struct net_device *ndev,
 517                          struct netvsc_device *net_dev,
 518                          struct netvsc_channel *nvchan,
 519                          void *data, u32 buflen)
 520 {
 521         struct net_device_context *net_device_ctx = netdev_priv(ndev);
 522         struct rndis_message *rndis_msg = data;
 523
 524         if (netif_msg_rx_status(net_device_ctx))
 525                 dump_rndis_message(ndev, rndis_msg);
 526
 527         /* Validate incoming rndis_message packet */
 528         if (buflen < RNDIS_HEADER_SIZE || rndis_msg->msg_len < RNDIS_HEADER_SIZE ||
 529             buflen < rndis_msg->msg_len) {
 530                 netdev_err(ndev, "Invalid rndis_msg (buflen: %u, msg_len: %u)\n",
 531                            buflen, rndis_msg->msg_len);
 532                 return NVSP_STAT_FAIL;
 533         }
 534
 535         switch (rndis_msg->ndis_msg_type) {
 536         case RNDIS_MSG_PACKET:
 537                 return rndis_filter_receive_data(ndev, net_dev, nvchan,
 538                                                  rndis_msg, buflen);
 539         case RNDIS_MSG_INIT_C:
 540         case RNDIS_MSG_QUERY_C:
 541         case RNDIS_MSG_SET_C:
 542                 /* completion msgs */
 543                 rndis_filter_receive_response(ndev, net_dev, rndis_msg);
 544                 break;
 545
 546         case RNDIS_MSG_INDICATE:
 547                 /* notification msgs */
 548                 netvsc_linkstatus_callback(ndev, rndis_msg);
 549                 break;
 550         default:
 551                 netdev_err(ndev,
 552                         "unhandled rndis message (type %u len %u)\n",
 553                            rndis_msg->ndis_msg_type,
 554                            rndis_msg->msg_len);
 555                 return NVSP_STAT_FAIL;
 556         }
 557
 558         return NVSP_STAT_SUCCESS;
 559 }
 560
 561 static int rndis_filter_query_device(struct rndis_device *dev,
 562                                      struct netvsc_device *nvdev,
 563                                      u32 oid, void *result, u32 *result_size)
 564 {
 565         struct rndis_request *request;
 566         u32 inresult_size = *result_size;
 567         struct rndis_query_request *query;
 568         struct rndis_query_complete *query_complete;
 569         int ret = 0;
 570
 571         if (!result)
 572                 return -EINVAL;
 573
 574         *result_size = 0;
 575         request = get_rndis_request(dev, RNDIS_MSG_QUERY,
 576                         RNDIS_MESSAGE_SIZE(struct rndis_query_request));
 577         if (!request) {
 578                 ret = -ENOMEM;
 579                 goto cleanup;
 580         }
 581
 582         /* Setup the rndis query */
 583         query = &request->request_msg.msg.query_req;
 584         query->oid = oid;
 585         query->info_buf_offset = sizeof(struct rndis_query_request);
 586         query->info_buflen = 0;
 587         query->dev_vc_handle = 0;
 588
 589         if (oid == OID_TCP_OFFLOAD_HARDWARE_CAPABILITIES) {
 590                 struct ndis_offload *hwcaps;
 591                 u32 nvsp_version = nvdev->nvsp_version;
 592                 u8 ndis_rev;
 593                 size_t size;
 594
 595                 if (nvsp_version >= NVSP_PROTOCOL_VERSION_5) {
 596                         ndis_rev = NDIS_OFFLOAD_PARAMETERS_REVISION_3;
 597                         size = NDIS_OFFLOAD_SIZE;
 598                 } else if (nvsp_version >= NVSP_PROTOCOL_VERSION_4) {
 599                         ndis_rev = NDIS_OFFLOAD_PARAMETERS_REVISION_2;
 600                         size = NDIS_OFFLOAD_SIZE_6_1;
 601                 } else {
 602                         ndis_rev = NDIS_OFFLOAD_PARAMETERS_REVISION_1;
 603                         size = NDIS_OFFLOAD_SIZE_6_0;
 604                 }
 605
 606                 request->request_msg.msg_len += size;
 607                 query->info_buflen = size;
 608                 hwcaps = (struct ndis_offload *)
 609                         ((unsigned long)query + query->info_buf_offset);
 610
 611                 hwcaps->header.type = NDIS_OBJECT_TYPE_OFFLOAD;
 612                 hwcaps->header.revision = ndis_rev;
 613                 hwcaps->header.size = size;
 614
 615         } else if (oid == OID_GEN_RECEIVE_SCALE_CAPABILITIES) {
 616                 struct ndis_recv_scale_cap *cap;
 617
 618                 request->request_msg.msg_len +=
 619                         sizeof(struct ndis_recv_scale_cap);
 620                 query->info_buflen = sizeof(struct ndis_recv_scale_cap);
 621                 cap = (struct ndis_recv_scale_cap *)((unsigned long)query +
 622                                                      query->info_buf_offset);
 623                 cap->hdr.type = NDIS_OBJECT_TYPE_RSS_CAPABILITIES;
 624                 cap->hdr.rev = NDIS_RECEIVE_SCALE_CAPABILITIES_REVISION_2;
 625                 cap->hdr.size = sizeof(struct ndis_recv_scale_cap);
 626         }
 627
 628         ret = rndis_filter_send_request(dev, request);
 629         if (ret != 0)
 630                 goto cleanup;
 631
 632         wait_for_completion(&request->wait_event);
 633
 634         /* Copy the response back */
 635         query_complete = &request->response_msg.msg.query_complete;
 636
 637         if (query_complete->info_buflen > inresult_size) {
 638                 ret = -1;
 639                 goto cleanup;
 640         }
 641
 642         memcpy(result,
 643                (void *)((unsigned long)query_complete +
 644                          query_complete->info_buf_offset),
 645                query_complete->info_buflen);
 646
 647         *result_size = query_complete->info_buflen;
 648
 649 cleanup:
 650         if (request)
 651                 put_rndis_request(dev, request);
 652
 653         return ret;
 654 }
 655
 656 /* Get the hardware offload capabilities */
 657 static int
 658 rndis_query_hwcaps(struct rndis_device *dev, struct netvsc_device *net_device,
 659                    struct ndis_offload *caps)
 660 {
 661         u32 caps_len = sizeof(*caps);
 662         int ret;
 663
 664         memset(caps, 0, sizeof(*caps));
 665
 666         ret = rndis_filter_query_device(dev, net_device,
 667                                         OID_TCP_OFFLOAD_HARDWARE_CAPABILITIES,
 668                                         caps, &caps_len);
 669         if (ret)
 670                 return ret;
 671
 672         if (caps->header.type != NDIS_OBJECT_TYPE_OFFLOAD) {
 673                 netdev_warn(dev->ndev, "invalid NDIS objtype %#x\n",
 674                             caps->header.type);
 675                 return -EINVAL;
 676         }
 677
 678         if (caps->header.revision < NDIS_OFFLOAD_PARAMETERS_REVISION_1) {
 679                 netdev_warn(dev->ndev, "invalid NDIS objrev %x\n",
 680                             caps->header.revision);
 681                 return -EINVAL;
 682         }
 683
 684         if (caps->header.size > caps_len ||
 685             caps->header.size < NDIS_OFFLOAD_SIZE_6_0) {
 686                 netdev_warn(dev->ndev,
 687                             "invalid NDIS objsize %u, data size %u\n",
 688                             caps->header.size, caps_len);
 689                 return -EINVAL;
 690         }
 691
 692         return 0;
 693 }
 694
 695 static int rndis_filter_query_device_mac(struct rndis_device *dev,
 696                                          struct netvsc_device *net_device)
 697 {
 698         u32 size = ETH_ALEN;
 699
 700         return rndis_filter_query_device(dev, net_device,
 701                                       RNDIS_OID_802_3_PERMANENT_ADDRESS,
 702                                       dev->hw_mac_adr, &size);
 703 }
 704
 705 #define NWADR_STR "NetworkAddress"
 706 #define NWADR_STRLEN 14
 707
 708 int rndis_filter_set_device_mac(struct netvsc_device *nvdev,
 709                                 const char *mac)
 710 {
 711         struct rndis_device *rdev = nvdev->extension;
 712         struct rndis_request *request;
 713         struct rndis_set_request *set;
 714         struct rndis_config_parameter_info *cpi;
 715         wchar_t *cfg_nwadr, *cfg_mac;
 716         struct rndis_set_complete *set_complete;
 717         char macstr[2*ETH_ALEN+1];
 718         u32 extlen = sizeof(struct rndis_config_parameter_info) +
 719                 2*NWADR_STRLEN + 4*ETH_ALEN;
 720         int ret;
 721
 722         request = get_rndis_request(rdev, RNDIS_MSG_SET,
 723                 RNDIS_MESSAGE_SIZE(struct rndis_set_request) + extlen);
 724         if (!request)
 725                 return -ENOMEM;
 726
 727         set = &request->request_msg.msg.set_req;
 728         set->oid = RNDIS_OID_GEN_RNDIS_CONFIG_PARAMETER;
 729         set->info_buflen = extlen;
 730         set->info_buf_offset = sizeof(struct rndis_set_request);
 731         set->dev_vc_handle = 0;
 732
 733         cpi = (struct rndis_config_parameter_info *)((ulong)set +
 734                 set->info_buf_offset);
 735         cpi->parameter_name_offset =
 736                 sizeof(struct rndis_config_parameter_info);
 737         /* Multiply by 2 because host needs 2 bytes (utf16) for each char */
 738         cpi->parameter_name_length = 2*NWADR_STRLEN;
 739         cpi->parameter_type = RNDIS_CONFIG_PARAM_TYPE_STRING;
 740         cpi->parameter_value_offset =
 741                 cpi->parameter_name_offset + cpi->parameter_name_length;
 742         /* Multiply by 4 because each MAC byte displayed as 2 utf16 chars */
 743         cpi->parameter_value_length = 4*ETH_ALEN;
 744
 745         cfg_nwadr = (wchar_t *)((ulong)cpi + cpi->parameter_name_offset);
 746         cfg_mac = (wchar_t *)((ulong)cpi + cpi->parameter_value_offset);
 747         ret = utf8s_to_utf16s(NWADR_STR, NWADR_STRLEN, UTF16_HOST_ENDIAN,
 748                               cfg_nwadr, NWADR_STRLEN);
 749         if (ret < 0)
 750                 goto cleanup;
 751         snprintf(macstr, 2*ETH_ALEN+1, "%pm", mac);
 752         ret = utf8s_to_utf16s(macstr, 2*ETH_ALEN, UTF16_HOST_ENDIAN,
 753                               cfg_mac, 2*ETH_ALEN);
 754         if (ret < 0)
 755                 goto cleanup;
 756
 757         ret = rndis_filter_send_request(rdev, request);
 758         if (ret != 0)
 759                 goto cleanup;
 760
 761         wait_for_completion(&request->wait_event);
 762
 763         set_complete = &request->response_msg.msg.set_complete;
 764         if (set_complete->status != RNDIS_STATUS_SUCCESS)
 765                 ret = -EIO;
 766
 767 cleanup:
 768         put_rndis_request(rdev, request);
 769         return ret;
 770 }
 771
 772 int
 773 rndis_filter_set_offload_params(struct net_device *ndev,
 774                                 struct netvsc_device *nvdev,
 775                                 struct ndis_offload_params *req_offloads)
 776 {
 777         struct rndis_device *rdev = nvdev->extension;
 778         struct rndis_request *request;
 779         struct rndis_set_request *set;
 780         struct ndis_offload_params *offload_params;
 781         struct rndis_set_complete *set_complete;
 782         u32 extlen = sizeof(struct ndis_offload_params);
 783         int ret;
 784         u32 vsp_version = nvdev->nvsp_version;
 785
 786         if (vsp_version <= NVSP_PROTOCOL_VERSION_4) {
 787                 extlen = VERSION_4_OFFLOAD_SIZE;
 788                 /* On NVSP_PROTOCOL_VERSION_4 and below, we do not support
 789                  * UDP checksum offload.
 790                  */
 791                 req_offloads->udp_ip_v4_csum = 0;
 792                 req_offloads->udp_ip_v6_csum = 0;
 793         }
 794
 795         request = get_rndis_request(rdev, RNDIS_MSG_SET,
 796                 RNDIS_MESSAGE_SIZE(struct rndis_set_request) + extlen);
 797         if (!request)
 798                 return -ENOMEM;
 799
 800         set = &request->request_msg.msg.set_req;
 801         set->oid = OID_TCP_OFFLOAD_PARAMETERS;
 802         set->info_buflen = extlen;
 803         set->info_buf_offset = sizeof(struct rndis_set_request);
 804         set->dev_vc_handle = 0;
 805
 806         offload_params = (struct ndis_offload_params *)((ulong)set +
 807                                 set->info_buf_offset);
 808         *offload_params = *req_offloads;
 809         offload_params->header.type = NDIS_OBJECT_TYPE_DEFAULT;
 810         offload_params->header.revision = NDIS_OFFLOAD_PARAMETERS_REVISION_3;
 811         offload_params->header.size = extlen;
 812
 813         ret = rndis_filter_send_request(rdev, request);
 814         if (ret != 0)
 815                 goto cleanup;
 816
 817         wait_for_completion(&request->wait_event);
 818         set_complete = &request->response_msg.msg.set_complete;
 819         if (set_complete->status != RNDIS_STATUS_SUCCESS) {
 820                 netdev_err(ndev, "Fail to set offload on host side:0x%x\n",
 821                            set_complete->status);
 822                 ret = -EINVAL;
 823         }
 824
 825 cleanup:
 826         put_rndis_request(rdev, request);
 827         return ret;
 828 }
 829
 830 static int rndis_set_rss_param_msg(struct rndis_device *rdev,
 831                                    const u8 *rss_key, u16 flag)
 832 {
 833         struct net_device *ndev = rdev->ndev;
 834         struct net_device_context *ndc = netdev_priv(ndev);
 835         struct rndis_request *request;
 836         struct rndis_set_request *set;
 837         struct rndis_set_complete *set_complete;
 838         u32 extlen = sizeof(struct ndis_recv_scale_param) +
 839                      4 * ITAB_NUM + NETVSC_HASH_KEYLEN;
 840         struct ndis_recv_scale_param *rssp;
 841         u32 *itab;
 842         u8 *keyp;
 843         int i, ret;
 844
 845         request = get_rndis_request(
 846                         rdev, RNDIS_MSG_SET,
 847                         RNDIS_MESSAGE_SIZE(struct rndis_set_request) + extlen);
 848         if (!request)
 849                 return -ENOMEM;
 850
 851         set = &request->request_msg.msg.set_req;
 852         set->oid = OID_GEN_RECEIVE_SCALE_PARAMETERS;
 853         set->info_buflen = extlen;
 854         set->info_buf_offset = sizeof(struct rndis_set_request);
 855         set->dev_vc_handle = 0;
 856
 857         rssp = (struct ndis_recv_scale_param *)(set + 1);
 858         rssp->hdr.type = NDIS_OBJECT_TYPE_RSS_PARAMETERS;
 859         rssp->hdr.rev = NDIS_RECEIVE_SCALE_PARAMETERS_REVISION_2;
 860         rssp->hdr.size = sizeof(struct ndis_recv_scale_param);
 861         rssp->flag = flag;
 862         rssp->hashinfo = NDIS_HASH_FUNC_TOEPLITZ | NDIS_HASH_IPV4 |
 863                          NDIS_HASH_TCP_IPV4 | NDIS_HASH_IPV6 |
 864                          NDIS_HASH_TCP_IPV6;
 865         rssp->indirect_tabsize = 4*ITAB_NUM;
 866         rssp->indirect_taboffset = sizeof(struct ndis_recv_scale_param);
 867         rssp->hashkey_size = NETVSC_HASH_KEYLEN;
 868         rssp->hashkey_offset = rssp->indirect_taboffset +
 869                                rssp->indirect_tabsize;
 870
 871         /* Set indirection table entries */
 872         itab = (u32 *)(rssp + 1);
 873         for (i = 0; i < ITAB_NUM; i++)
 874                 itab[i] = ndc->rx_table[i];
 875
 876         /* Set hask key values */
 877         keyp = (u8 *)((unsigned long)rssp + rssp->hashkey_offset);
 878         memcpy(keyp, rss_key, NETVSC_HASH_KEYLEN);
 879
 880         ret = rndis_filter_send_request(rdev, request);
 881         if (ret != 0)
 882                 goto cleanup;
 883
 884         wait_for_completion(&request->wait_event);
 885         set_complete = &request->response_msg.msg.set_complete;
 886         if (set_complete->status == RNDIS_STATUS_SUCCESS) {
 887                 if (!(flag & NDIS_RSS_PARAM_FLAG_DISABLE_RSS) &&
 888                     !(flag & NDIS_RSS_PARAM_FLAG_HASH_KEY_UNCHANGED))
 889                         memcpy(rdev->rss_key, rss_key, NETVSC_HASH_KEYLEN);
 890
 891         } else {
 892                 netdev_err(ndev, "Fail to set RSS parameters:0x%x\n",
 893                            set_complete->status);
 894                 ret = -EINVAL;
 895         }
 896
 897 cleanup:
 898         put_rndis_request(rdev, request);
 899         return ret;
 900 }
 901
 902 int rndis_filter_set_rss_param(struct rndis_device *rdev,
 903                                const u8 *rss_key)
 904 {
 905         /* Disable RSS before change */
 906         rndis_set_rss_param_msg(rdev, rss_key,
 907                                 NDIS_RSS_PARAM_FLAG_DISABLE_RSS);
 908
 909         return rndis_set_rss_param_msg(rdev, rss_key, 0);
 910 }
 911
 912 static int rndis_filter_query_device_link_status(struct rndis_device *dev,
 913                                                  struct netvsc_device *net_device)
 914 {
 915         u32 size = sizeof(u32);
 916         u32 link_status;
 917
 918         return rndis_filter_query_device(dev, net_device,
 919                                          RNDIS_OID_GEN_MEDIA_CONNECT_STATUS,
 920                                          &link_status, &size);
 921 }
 922
 923 static int rndis_filter_query_link_speed(struct rndis_device *dev,
 924                                          struct netvsc_device *net_device)
 925 {
 926         u32 size = sizeof(u32);
 927         u32 link_speed;
 928         struct net_device_context *ndc;
 929         int ret;
 930
 931         ret = rndis_filter_query_device(dev, net_device,
 932                                         RNDIS_OID_GEN_LINK_SPEED,
 933                                         &link_speed, &size);
 934
 935         if (!ret) {
 936                 ndc = netdev_priv(dev->ndev);
 937
 938                 /* The link speed reported from host is in 100bps unit, so
 939                  * we convert it to Mbps here.
 940                  */
 941                 ndc->speed = link_speed / 10000;
 942         }
 943
 944         return ret;
 945 }
 946
 947 static int rndis_filter_set_packet_filter(struct rndis_device *dev,
 948                                           u32 new_filter)
 949 {
 950         struct rndis_request *request;
 951         struct rndis_set_request *set;
 952         int ret;
 953
 954         if (dev->filter == new_filter)
 955                 return 0;
 956
 957         request = get_rndis_request(dev, RNDIS_MSG_SET,
 958                         RNDIS_MESSAGE_SIZE(struct rndis_set_request) +
 959                         sizeof(u32));
 960         if (!request)
 961                 return -ENOMEM;
 962
 963         /* Setup the rndis set */
 964         set = &request->request_msg.msg.set_req;
 965         set->oid = RNDIS_OID_GEN_CURRENT_PACKET_FILTER;
 966         set->info_buflen = sizeof(u32);
 967         set->info_buf_offset = sizeof(struct rndis_set_request);
 968
 969         memcpy((void *)(unsigned long)set + sizeof(struct rndis_set_request),
 970                &new_filter, sizeof(u32));
 971
 972         ret = rndis_filter_send_request(dev, request);
 973         if (ret == 0) {
 974                 wait_for_completion(&request->wait_event);
 975                 dev->filter = new_filter;
 976         }
 977
 978         put_rndis_request(dev, request);
 979
 980         return ret;
 981 }
 982
 983 static void rndis_set_multicast(struct work_struct *w)
 984 {
 985         struct rndis_device *rdev
 986                 = container_of(w, struct rndis_device, mcast_work);
 987         u32 filter = NDIS_PACKET_TYPE_DIRECTED;
 988         unsigned int flags = rdev->ndev->flags;
 989
 990         if (flags & IFF_PROMISC) {
 991                 filter = NDIS_PACKET_TYPE_PROMISCUOUS;
 992         } else {
 993                 if (!netdev_mc_empty(rdev->ndev) || (flags & IFF_ALLMULTI))
 994                         filter |= NDIS_PACKET_TYPE_ALL_MULTICAST;
 995                 if (flags & IFF_BROADCAST)
 996                         filter |= NDIS_PACKET_TYPE_BROADCAST;
 997         }
 998
 999         rndis_filter_set_packet_filter(rdev, filter);
1000 }
1001
1002 void rndis_filter_update(struct netvsc_device *nvdev)
1003 {
1004         struct rndis_device *rdev = nvdev->extension;
1005
1006         schedule_work(&rdev->mcast_work);
1007 }
1008
1009 static int rndis_filter_init_device(struct rndis_device *dev,
1010                                     struct netvsc_device *nvdev)
1011 {
1012         struct rndis_request *request;
1013         struct rndis_initialize_request *init;
1014         struct rndis_initialize_complete *init_complete;
1015         u32 status;
1016         int ret;
1017
1018         request = get_rndis_request(dev, RNDIS_MSG_INIT,
1019                         RNDIS_MESSAGE_SIZE(struct rndis_initialize_request));
1020         if (!request) {
1021                 ret = -ENOMEM;
1022                 goto cleanup;
1023         }
1024
1025         /* Setup the rndis set */
1026         init = &request->request_msg.msg.init_req;
1027         init->major_ver = RNDIS_MAJOR_VERSION;
1028         init->minor_ver = RNDIS_MINOR_VERSION;
1029         init->max_xfer_size = 0x4000;
1030
1031         dev->state = RNDIS_DEV_INITIALIZING;
1032
1033         ret = rndis_filter_send_request(dev, request);
1034         if (ret != 0) {
1035                 dev->state = RNDIS_DEV_UNINITIALIZED;
1036                 goto cleanup;
1037         }
1038
1039         wait_for_completion(&request->wait_event);
1040
1041         init_complete = &request->response_msg.msg.init_complete;
1042         status = init_complete->status;
1043         if (status == RNDIS_STATUS_SUCCESS) {
1044                 dev->state = RNDIS_DEV_INITIALIZED;
1045                 nvdev->max_pkt = init_complete->max_pkt_per_msg;
1046                 nvdev->pkt_align = 1 << init_complete->pkt_alignment_factor;
1047                 ret = 0;
1048         } else {
1049                 dev->state = RNDIS_DEV_UNINITIALIZED;
1050                 ret = -EINVAL;
1051         }
1052
1053 cleanup:
1054         if (request)
1055                 put_rndis_request(dev, request);
1056
1057         return ret;
1058 }
1059
1060 static bool netvsc_device_idle(const struct netvsc_device *nvdev)
1061 {
1062         int i;
1063
1064         for (i = 0; i < nvdev->num_chn; i++) {
1065                 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1066
1067                 if (nvchan->mrc.first != nvchan->mrc.next)
1068                         return false;
1069
1070                 if (atomic_read(&nvchan->queue_sends) > 0)
1071                         return false;
1072         }
1073
1074         return true;
1075 }
1076
1077 static void rndis_filter_halt_device(struct netvsc_device *nvdev,
1078                                      struct rndis_device *dev)
1079 {
1080         struct rndis_request *request;
1081         struct rndis_halt_request *halt;
1082
1083         /* Attempt to do a rndis device halt */
1084         request = get_rndis_request(dev, RNDIS_MSG_HALT,
1085                                 RNDIS_MESSAGE_SIZE(struct rndis_halt_request));
1086         if (!request)
1087                 goto cleanup;
1088
1089         /* Setup the rndis set */
1090         halt = &request->request_msg.msg.halt_req;
1091         halt->req_id = atomic_inc_return(&dev->new_req_id);
1092
1093         /* Ignore return since this msg is optional. */
1094         rndis_filter_send_request(dev, request);
1095
1096         dev->state = RNDIS_DEV_UNINITIALIZED;
1097
1098 cleanup:
1099         nvdev->destroy = true;
1100
1101         /* Force flag to be ordered before waiting */
1102         wmb();
1103
1104         /* Wait for all send completions */
1105         wait_event(nvdev->wait_drain, netvsc_device_idle(nvdev));
1106
1107         if (request)
1108                 put_rndis_request(dev, request);
1109 }
1110
1111 static int rndis_filter_open_device(struct rndis_device *dev)
1112 {
1113         int ret;
1114
1115         if (dev->state != RNDIS_DEV_INITIALIZED)
1116                 return 0;
1117
1118         ret = rndis_filter_set_packet_filter(dev,
1119                                          NDIS_PACKET_TYPE_BROADCAST |
1120                                          NDIS_PACKET_TYPE_ALL_MULTICAST |
1121                                          NDIS_PACKET_TYPE_DIRECTED);
1122         if (ret == 0)
1123                 dev->state = RNDIS_DEV_DATAINITIALIZED;
1124
1125         return ret;
1126 }
1127
1128 static int rndis_filter_close_device(struct rndis_device *dev)
1129 {
1130         int ret;
1131
1132         if (dev->state != RNDIS_DEV_DATAINITIALIZED)
1133                 return 0;
1134
1135         /* Make sure rndis_set_multicast doesn't re-enable filter! */
1136         cancel_work_sync(&dev->mcast_work);
1137
1138         ret = rndis_filter_set_packet_filter(dev, 0);
1139         if (ret == -ENODEV)
1140                 ret = 0;
1141
1142         if (ret == 0)
1143                 dev->state = RNDIS_DEV_INITIALIZED;
1144
1145         return ret;
1146 }
1147
1148 static void netvsc_sc_open(struct vmbus_channel *new_sc)
1149 {
1150         struct net_device *ndev =
1151                 hv_get_drvdata(new_sc->primary_channel->device_obj);
1152         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1153         struct netvsc_device *nvscdev;
1154         u16 chn_index = new_sc->offermsg.offer.sub_channel_index;
1155         struct netvsc_channel *nvchan;
1156         int ret;
1157
1158         /* This is safe because this callback only happens when
1159          * new device is being setup and waiting on the channel_init_wait.
1160          */
1161         nvscdev = rcu_dereference_raw(ndev_ctx->nvdev);
1162         if (!nvscdev || chn_index >= nvscdev->num_chn)
1163                 return;
1164
1165         nvchan = nvscdev->chan_table + chn_index;
1166
1167         /* Because the device uses NAPI, all the interrupt batching and
1168          * control is done via Net softirq, not the channel handling
1169          */
1170         set_channel_read_mode(new_sc, HV_CALL_ISR);
1171
1172         /* Set the channel before opening.*/
1173         nvchan->channel = new_sc;
1174
1175         ret = vmbus_open(new_sc, netvsc_ring_bytes,
1176                          netvsc_ring_bytes, NULL, 0,
1177                          netvsc_channel_cb, nvchan);
1178         if (ret == 0)
1179                 napi_enable(&nvchan->napi);
1180         else
1181                 netdev_notice(ndev, "sub channel open failed: %d\n", ret);
1182
1183         if (atomic_inc_return(&nvscdev->open_chn) == nvscdev->num_chn)
1184                 wake_up(&nvscdev->subchan_open);
1185 }
1186
1187 /* Open sub-channels after completing the handling of the device probe.
1188  * This breaks overlap of processing the host message for the
1189  * new primary channel with the initialization of sub-channels.
1190  */
1191 int rndis_set_subchannel(struct net_device *ndev,
1192                          struct netvsc_device *nvdev,
1193                          struct netvsc_device_info *dev_info)
1194 {
1195         struct nvsp_message *init_packet = &nvdev->channel_init_pkt;
1196         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1197         struct hv_device *hv_dev = ndev_ctx->device_ctx;
1198         struct rndis_device *rdev = nvdev->extension;
1199         int i, ret;
1200
1201         ASSERT_RTNL();
1202
1203         memset(init_packet, 0, sizeof(struct nvsp_message));
1204         init_packet->hdr.msg_type = NVSP_MSG5_TYPE_SUBCHANNEL;
1205         init_packet->msg.v5_msg.subchn_req.op = NVSP_SUBCHANNEL_ALLOCATE;
1206         init_packet->msg.v5_msg.subchn_req.num_subchannels =
1207                                                 nvdev->num_chn - 1;
1208         trace_nvsp_send(ndev, init_packet);
1209
1210         ret = vmbus_sendpacket(hv_dev->channel, init_packet,
1211                                sizeof(struct nvsp_message),
1212                                (unsigned long)init_packet,
1213                                VM_PKT_DATA_INBAND,
1214                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1215         if (ret) {
1216                 netdev_err(ndev, "sub channel allocate send failed: %d\n", ret);
1217                 return ret;
1218         }
1219
1220         wait_for_completion(&nvdev->channel_init_wait);
1221         if (init_packet->msg.v5_msg.subchn_comp.status != NVSP_STAT_SUCCESS) {
1222                 netdev_err(ndev, "sub channel request failed\n");
1223                 return -EIO;
1224         }
1225
1226         nvdev->num_chn = 1 +
1227                 init_packet->msg.v5_msg.subchn_comp.num_subchannels;
1228
1229         /* wait for all sub channels to open */
1230         wait_event(nvdev->subchan_open,
1231                    atomic_read(&nvdev->open_chn) == nvdev->num_chn);
1232
1233         for (i = 0; i < VRSS_SEND_TAB_SIZE; i++)
1234                 ndev_ctx->tx_table[i] = i % nvdev->num_chn;
1235
1236         /* ignore failures from setting rss parameters, still have channels */
1237         if (dev_info)
1238                 rndis_filter_set_rss_param(rdev, dev_info->rss_key);
1239         else
1240                 rndis_filter_set_rss_param(rdev, netvsc_hash_key);
1241
1242         netif_set_real_num_tx_queues(ndev, nvdev->num_chn);
1243         netif_set_real_num_rx_queues(ndev, nvdev->num_chn);
1244
1245         return 0;
1246 }
1247
1248 static int rndis_netdev_set_hwcaps(struct rndis_device *rndis_device,
1249                                    struct netvsc_device *nvdev)
1250 {
1251         struct net_device *net = rndis_device->ndev;
1252         struct net_device_context *net_device_ctx = netdev_priv(net);
1253         struct ndis_offload hwcaps;
1254         struct ndis_offload_params offloads;
1255         unsigned int gso_max_size = GSO_MAX_SIZE;
1256         int ret;
1257
1258         /* Find HW offload capabilities */
1259         ret = rndis_query_hwcaps(rndis_device, nvdev, &hwcaps);
1260         if (ret != 0)
1261                 return ret;
1262
1263         /* A value of zero means "no change"; now turn on what we want. */
1264         memset(&offloads, 0, sizeof(struct ndis_offload_params));
1265
1266         /* Linux does not care about IP checksum, always does in kernel */
1267         offloads.ip_v4_csum = NDIS_OFFLOAD_PARAMETERS_TX_RX_DISABLED;
1268
1269         /* Reset previously set hw_features flags */
1270         net->hw_features &= ~NETVSC_SUPPORTED_HW_FEATURES;
1271         net_device_ctx->tx_checksum_mask = 0;
1272
1273         /* Compute tx offload settings based on hw capabilities */
1274         net->hw_features |= NETIF_F_RXCSUM;
1275         net->hw_features |= NETIF_F_SG;
1276         net->hw_features |= NETIF_F_RXHASH;
1277
1278         if ((hwcaps.csum.ip4_txcsum & NDIS_TXCSUM_ALL_TCP4) == NDIS_TXCSUM_ALL_TCP4) {
1279                 /* Can checksum TCP */
1280                 net->hw_features |= NETIF_F_IP_CSUM;
1281                 net_device_ctx->tx_checksum_mask |= TRANSPORT_INFO_IPV4_TCP;
1282
1283                 offloads.tcp_ip_v4_csum = NDIS_OFFLOAD_PARAMETERS_TX_RX_ENABLED;
1284
1285                 if (hwcaps.lsov2.ip4_encap & NDIS_OFFLOAD_ENCAP_8023) {
1286                         offloads.lso_v2_ipv4 = NDIS_OFFLOAD_PARAMETERS_LSOV2_ENABLED;
1287                         net->hw_features |= NETIF_F_TSO;
1288
1289                         if (hwcaps.lsov2.ip4_maxsz < gso_max_size)
1290                                 gso_max_size = hwcaps.lsov2.ip4_maxsz;
1291                 }
1292
1293                 if (hwcaps.csum.ip4_txcsum & NDIS_TXCSUM_CAP_UDP4) {
1294                         offloads.udp_ip_v4_csum = NDIS_OFFLOAD_PARAMETERS_TX_RX_ENABLED;
1295                         net_device_ctx->tx_checksum_mask |= TRANSPORT_INFO_IPV4_UDP;
1296                 }
1297         }
1298
1299         if ((hwcaps.csum.ip6_txcsum & NDIS_TXCSUM_ALL_TCP6) == NDIS_TXCSUM_ALL_TCP6) {
1300                 net->hw_features |= NETIF_F_IPV6_CSUM;
1301
1302                 offloads.tcp_ip_v6_csum = NDIS_OFFLOAD_PARAMETERS_TX_RX_ENABLED;
1303                 net_device_ctx->tx_checksum_mask |= TRANSPORT_INFO_IPV6_TCP;
1304
1305                 if ((hwcaps.lsov2.ip6_encap & NDIS_OFFLOAD_ENCAP_8023) &&
1306                     (hwcaps.lsov2.ip6_opts & NDIS_LSOV2_CAP_IP6) == NDIS_LSOV2_CAP_IP6) {
1307                         offloads.lso_v2_ipv6 = NDIS_OFFLOAD_PARAMETERS_LSOV2_ENABLED;
1308                         net->hw_features |= NETIF_F_TSO6;
1309
1310                         if (hwcaps.lsov2.ip6_maxsz < gso_max_size)
1311                                 gso_max_size = hwcaps.lsov2.ip6_maxsz;
1312                 }
1313
1314                 if (hwcaps.csum.ip6_txcsum & NDIS_TXCSUM_CAP_UDP6) {
1315                         offloads.udp_ip_v6_csum = NDIS_OFFLOAD_PARAMETERS_TX_RX_ENABLED;
1316                         net_device_ctx->tx_checksum_mask |= TRANSPORT_INFO_IPV6_UDP;
1317                 }
1318         }
1319
1320         if (hwcaps.rsc.ip4 && hwcaps.rsc.ip6) {
1321                 net->hw_features |= NETIF_F_LRO;
1322
1323                 if (net->features & NETIF_F_LRO) {
1324                         offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1325                         offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1326                 } else {
1327                         offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1328                         offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1329                 }
1330         }
1331
1332         /* In case some hw_features disappeared we need to remove them from
1333          * net->features list as they're no longer supported.
1334          */
1335         net->features &= ~NETVSC_SUPPORTED_HW_FEATURES | net->hw_features;
1336
1337         netif_set_gso_max_size(net, gso_max_size);
1338
1339         ret = rndis_filter_set_offload_params(net, nvdev, &offloads);
1340
1341         return ret;
1342 }
1343
1344 static void rndis_get_friendly_name(struct net_device *net,
1345                                     struct rndis_device *rndis_device,
1346                                     struct netvsc_device *net_device)
1347 {
1348         ucs2_char_t wname[256];
1349         unsigned long len;
1350         u8 ifalias[256];
1351         u32 size;
1352
1353         size = sizeof(wname);
1354         if (rndis_filter_query_device(rndis_device, net_device,
1355                                       RNDIS_OID_GEN_FRIENDLY_NAME,
1356                                       wname, &size) != 0)
1357                 return; /* ignore if host does not support */
1358
1359         if (size == 0)
1360                 return; /* name not set */
1361
1362         /* Convert Windows Unicode string to UTF-8 */
1363         len = ucs2_as_utf8(ifalias, wname, sizeof(ifalias));
1364
1365         /* ignore the default value from host */
1366         if (strcmp(ifalias, "Network Adapter") != 0)
1367                 dev_set_alias(net, ifalias, len);
1368 }
1369
1370 struct netvsc_device *rndis_filter_device_add(struct hv_device *dev,
1371                                       struct netvsc_device_info *device_info)
1372 {
1373         struct net_device *net = hv_get_drvdata(dev);
1374         struct net_device_context *ndc = netdev_priv(net);
1375         struct netvsc_device *net_device;
1376         struct rndis_device *rndis_device;
1377         struct ndis_recv_scale_cap rsscap;
1378         u32 rsscap_size = sizeof(struct ndis_recv_scale_cap);
1379         u32 mtu, size;
1380         u32 num_possible_rss_qs;
1381         int i, ret;
1382
1383         rndis_device = get_rndis_device();
1384         if (!rndis_device)
1385                 return ERR_PTR(-ENODEV);
1386
1387         /* Let the inner driver handle this first to create the netvsc channel
1388          * NOTE! Once the channel is created, we may get a receive callback
1389          * (RndisFilterOnReceive()) before this call is completed
1390          */
1391         net_device = netvsc_device_add(dev, device_info);
1392         if (IS_ERR(net_device)) {
1393                 kfree(rndis_device);
1394                 return net_device;
1395         }
1396
1397         /* Initialize the rndis device */
1398         net_device->max_chn = 1;
1399         net_device->num_chn = 1;
1400
1401         net_device->extension = rndis_device;
1402         rndis_device->ndev = net;
1403
1404         /* Send the rndis initialization message */
1405         ret = rndis_filter_init_device(rndis_device, net_device);
1406         if (ret != 0)
1407                 goto err_dev_remv;
1408
1409         /* Get the MTU from the host */
1410         size = sizeof(u32);
1411         ret = rndis_filter_query_device(rndis_device, net_device,
1412                                         RNDIS_OID_GEN_MAXIMUM_FRAME_SIZE,
1413                                         &mtu, &size);
1414         if (ret == 0 && size == sizeof(u32) && mtu < net->mtu)
1415                 net->mtu = mtu;
1416
1417         /* Get the mac address */
1418         ret = rndis_filter_query_device_mac(rndis_device, net_device);
1419         if (ret != 0)
1420                 goto err_dev_remv;
1421
1422         memcpy(device_info->mac_adr, rndis_device->hw_mac_adr, ETH_ALEN);
1423
1424         /* Get friendly name as ifalias*/
1425         if (!net->ifalias)
1426                 rndis_get_friendly_name(net, rndis_device, net_device);
1427
1428         /* Query and set hardware capabilities */
1429         ret = rndis_netdev_set_hwcaps(rndis_device, net_device);
1430         if (ret != 0)
1431                 goto err_dev_remv;
1432
1433         rndis_filter_query_device_link_status(rndis_device, net_device);
1434
1435         netdev_dbg(net, "Device MAC %pM link state %s\n",
1436                    rndis_device->hw_mac_adr,
1437                    rndis_device->link_state ? "down" : "up");
1438
1439         if (net_device->nvsp_version < NVSP_PROTOCOL_VERSION_5)
1440                 goto out;
1441
1442         rndis_filter_query_link_speed(rndis_device, net_device);
1443
1444         /* vRSS setup */
1445         memset(&rsscap, 0, rsscap_size);
1446         ret = rndis_filter_query_device(rndis_device, net_device,
1447                                         OID_GEN_RECEIVE_SCALE_CAPABILITIES,
1448                                         &rsscap, &rsscap_size);
1449         if (ret || rsscap.num_recv_que < 2)
1450                 goto out;
1451
1452         /* This guarantees that num_possible_rss_qs <= num_online_cpus */
1453         num_possible_rss_qs = min_t(u32, num_online_cpus(),
1454                                     rsscap.num_recv_que);
1455
1456         net_device->max_chn = min_t(u32, VRSS_CHANNEL_MAX, num_possible_rss_qs);
1457
1458         /* We will use the given number of channels if available. */
1459         net_device->num_chn = min(net_device->max_chn, device_info->num_chn);
1460
1461         if (!netif_is_rxfh_configured(net)) {
1462                 for (i = 0; i < ITAB_NUM; i++)
1463                         ndc->rx_table[i] = ethtool_rxfh_indir_default(
1464                                                 i, net_device->num_chn);
1465         }
1466
1467         atomic_set(&net_device->open_chn, 1);
1468         vmbus_set_sc_create_callback(dev->channel, netvsc_sc_open);
1469
1470         for (i = 1; i < net_device->num_chn; i++) {
1471                 ret = netvsc_alloc_recv_comp_ring(net_device, i);
1472                 if (ret) {
1473                         while (--i != 0)
1474                                 vfree(net_device->chan_table[i].mrc.slots);
1475                         goto out;
1476                 }
1477         }
1478
1479         for (i = 1; i < net_device->num_chn; i++)
1480                 netif_napi_add(net, &net_device->chan_table[i].napi,
1481                                netvsc_poll, NAPI_POLL_WEIGHT);
1482
1483         return net_device;
1484
1485 out:
1486         /* setting up multiple channels failed */
1487         net_device->max_chn = 1;
1488         net_device->num_chn = 1;
1489         return net_device;
1490
1491 err_dev_remv:
1492         rndis_filter_device_remove(dev, net_device);
1493         return ERR_PTR(ret);
1494 }
1495
1496 void rndis_filter_device_remove(struct hv_device *dev,
1497                                 struct netvsc_device *net_dev)
1498 {
1499         struct rndis_device *rndis_dev = net_dev->extension;
1500
1501         /* Halt and release the rndis device */
1502         rndis_filter_halt_device(net_dev, rndis_dev);
1503
1504         netvsc_device_remove(dev);
1505 }
1506
1507 int rndis_filter_open(struct netvsc_device *nvdev)
1508 {
1509         if (!nvdev)
1510                 return -EINVAL;
1511
1512         return rndis_filter_open_device(nvdev->extension);
1513 }
1514
1515 int rndis_filter_close(struct netvsc_device *nvdev)
1516 {
1517         if (!nvdev)
1518                 return -EINVAL;
1519
1520         return rndis_filter_close_device(nvdev->extension);
1521 }