Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / net / hyperv / rndis_filter.c
diff --git a/kernel/drivers/net/hyperv/rndis_filter.c b/kernel/drivers/net/hyperv/rndis_filter.c
new file mode 100644 (file)
index 0000000..9118cea
--- /dev/null
@@ -0,0 +1,1192 @@
+/*
+ * Copyright (c) 2009, Microsoft Corporation.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+ * more details.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * this program; if not, see <http://www.gnu.org/licenses/>.
+ *
+ * Authors:
+ *   Haiyang Zhang <haiyangz@microsoft.com>
+ *   Hank Janssen  <hjanssen@microsoft.com>
+ */
+#include <linux/kernel.h>
+#include <linux/sched.h>
+#include <linux/wait.h>
+#include <linux/highmem.h>
+#include <linux/slab.h>
+#include <linux/io.h>
+#include <linux/if_ether.h>
+#include <linux/netdevice.h>
+#include <linux/if_vlan.h>
+#include <linux/nls.h>
+
+#include "hyperv_net.h"
+
+
+#define RNDIS_EXT_LEN PAGE_SIZE
+struct rndis_request {
+       struct list_head list_ent;
+       struct completion  wait_event;
+
+       struct rndis_message response_msg;
+       /*
+        * The buffer for extended info after the RNDIS response message. It's
+        * referenced based on the data offset in the RNDIS message. Its size
+        * is enough for current needs, and should be sufficient for the near
+        * future.
+        */
+       u8 response_ext[RNDIS_EXT_LEN];
+
+       /* Simplify allocation by having a netvsc packet inline */
+       struct hv_netvsc_packet pkt;
+
+       struct rndis_message request_msg;
+       /*
+        * The buffer for the extended info after the RNDIS request message.
+        * It is referenced and sized in a similar way as response_ext.
+        */
+       u8 request_ext[RNDIS_EXT_LEN];
+};
+
+static struct rndis_device *get_rndis_device(void)
+{
+       struct rndis_device *device;
+
+       device = kzalloc(sizeof(struct rndis_device), GFP_KERNEL);
+       if (!device)
+               return NULL;
+
+       spin_lock_init(&device->request_lock);
+
+       INIT_LIST_HEAD(&device->req_list);
+
+       device->state = RNDIS_DEV_UNINITIALIZED;
+
+       return device;
+}
+
+static struct rndis_request *get_rndis_request(struct rndis_device *dev,
+                                            u32 msg_type,
+                                            u32 msg_len)
+{
+       struct rndis_request *request;
+       struct rndis_message *rndis_msg;
+       struct rndis_set_request *set;
+       unsigned long flags;
+
+       request = kzalloc(sizeof(struct rndis_request), GFP_KERNEL);
+       if (!request)
+               return NULL;
+
+       init_completion(&request->wait_event);
+
+       rndis_msg = &request->request_msg;
+       rndis_msg->ndis_msg_type = msg_type;
+       rndis_msg->msg_len = msg_len;
+
+       request->pkt.q_idx = 0;
+
+       /*
+        * Set the request id. This field is always after the rndis header for
+        * request/response packet types so we just used the SetRequest as a
+        * template
+        */
+       set = &rndis_msg->msg.set_req;
+       set->req_id = atomic_inc_return(&dev->new_req_id);
+
+       /* Add to the request list */
+       spin_lock_irqsave(&dev->request_lock, flags);
+       list_add_tail(&request->list_ent, &dev->req_list);
+       spin_unlock_irqrestore(&dev->request_lock, flags);
+
+       return request;
+}
+
+static void put_rndis_request(struct rndis_device *dev,
+                           struct rndis_request *req)
+{
+       unsigned long flags;
+
+       spin_lock_irqsave(&dev->request_lock, flags);
+       list_del(&req->list_ent);
+       spin_unlock_irqrestore(&dev->request_lock, flags);
+
+       kfree(req);
+}
+
+static void dump_rndis_message(struct hv_device *hv_dev,
+                       struct rndis_message *rndis_msg)
+{
+       struct net_device *netdev;
+       struct netvsc_device *net_device;
+
+       net_device = hv_get_drvdata(hv_dev);
+       netdev = net_device->ndev;
+
+       switch (rndis_msg->ndis_msg_type) {
+       case RNDIS_MSG_PACKET:
+               netdev_dbg(netdev, "RNDIS_MSG_PACKET (len %u, "
+                          "data offset %u data len %u, # oob %u, "
+                          "oob offset %u, oob len %u, pkt offset %u, "
+                          "pkt len %u\n",
+                          rndis_msg->msg_len,
+                          rndis_msg->msg.pkt.data_offset,
+                          rndis_msg->msg.pkt.data_len,
+                          rndis_msg->msg.pkt.num_oob_data_elements,
+                          rndis_msg->msg.pkt.oob_data_offset,
+                          rndis_msg->msg.pkt.oob_data_len,
+                          rndis_msg->msg.pkt.per_pkt_info_offset,
+                          rndis_msg->msg.pkt.per_pkt_info_len);
+               break;
+
+       case RNDIS_MSG_INIT_C:
+               netdev_dbg(netdev, "RNDIS_MSG_INIT_C "
+                       "(len %u, id 0x%x, status 0x%x, major %d, minor %d, "
+                       "device flags %d, max xfer size 0x%x, max pkts %u, "
+                       "pkt aligned %u)\n",
+                       rndis_msg->msg_len,
+                       rndis_msg->msg.init_complete.req_id,
+                       rndis_msg->msg.init_complete.status,
+                       rndis_msg->msg.init_complete.major_ver,
+                       rndis_msg->msg.init_complete.minor_ver,
+                       rndis_msg->msg.init_complete.dev_flags,
+                       rndis_msg->msg.init_complete.max_xfer_size,
+                       rndis_msg->msg.init_complete.
+                          max_pkt_per_msg,
+                       rndis_msg->msg.init_complete.
+                          pkt_alignment_factor);
+               break;
+
+       case RNDIS_MSG_QUERY_C:
+               netdev_dbg(netdev, "RNDIS_MSG_QUERY_C "
+                       "(len %u, id 0x%x, status 0x%x, buf len %u, "
+                       "buf offset %u)\n",
+                       rndis_msg->msg_len,
+                       rndis_msg->msg.query_complete.req_id,
+                       rndis_msg->msg.query_complete.status,
+                       rndis_msg->msg.query_complete.
+                          info_buflen,
+                       rndis_msg->msg.query_complete.
+                          info_buf_offset);
+               break;
+
+       case RNDIS_MSG_SET_C:
+               netdev_dbg(netdev,
+                       "RNDIS_MSG_SET_C (len %u, id 0x%x, status 0x%x)\n",
+                       rndis_msg->msg_len,
+                       rndis_msg->msg.set_complete.req_id,
+                       rndis_msg->msg.set_complete.status);
+               break;
+
+       case RNDIS_MSG_INDICATE:
+               netdev_dbg(netdev, "RNDIS_MSG_INDICATE "
+                       "(len %u, status 0x%x, buf len %u, buf offset %u)\n",
+                       rndis_msg->msg_len,
+                       rndis_msg->msg.indicate_status.status,
+                       rndis_msg->msg.indicate_status.status_buflen,
+                       rndis_msg->msg.indicate_status.status_buf_offset);
+               break;
+
+       default:
+               netdev_dbg(netdev, "0x%x (len %u)\n",
+                       rndis_msg->ndis_msg_type,
+                       rndis_msg->msg_len);
+               break;
+       }
+}
+
+static int rndis_filter_send_request(struct rndis_device *dev,
+                                 struct rndis_request *req)
+{
+       int ret;
+       struct hv_netvsc_packet *packet;
+       struct hv_page_buffer page_buf[2];
+
+       /* Setup the packet to send it */
+       packet = &req->pkt;
+
+       packet->is_data_pkt = false;
+       packet->total_data_buflen = req->request_msg.msg_len;
+       packet->page_buf_cnt = 1;
+       packet->page_buf = page_buf;
+
+       packet->page_buf[0].pfn = virt_to_phys(&req->request_msg) >>
+                                       PAGE_SHIFT;
+       packet->page_buf[0].len = req->request_msg.msg_len;
+       packet->page_buf[0].offset =
+               (unsigned long)&req->request_msg & (PAGE_SIZE - 1);
+
+       /* Add one page_buf when request_msg crossing page boundary */
+       if (packet->page_buf[0].offset + packet->page_buf[0].len > PAGE_SIZE) {
+               packet->page_buf_cnt++;
+               packet->page_buf[0].len = PAGE_SIZE -
+                       packet->page_buf[0].offset;
+               packet->page_buf[1].pfn = virt_to_phys((void *)&req->request_msg
+                       + packet->page_buf[0].len) >> PAGE_SHIFT;
+               packet->page_buf[1].offset = 0;
+               packet->page_buf[1].len = req->request_msg.msg_len -
+                       packet->page_buf[0].len;
+       }
+
+       packet->send_completion = NULL;
+       packet->xmit_more = false;
+
+       ret = netvsc_send(dev->net_dev->dev, packet);
+       return ret;
+}
+
+static void rndis_set_link_state(struct rndis_device *rdev,
+                                struct rndis_request *request)
+{
+       u32 link_status;
+       struct rndis_query_complete *query_complete;
+
+       query_complete = &request->response_msg.msg.query_complete;
+
+       if (query_complete->status == RNDIS_STATUS_SUCCESS &&
+           query_complete->info_buflen == sizeof(u32)) {
+               memcpy(&link_status, (void *)((unsigned long)query_complete +
+                      query_complete->info_buf_offset), sizeof(u32));
+               rdev->link_state = link_status != 0;
+       }
+}
+
+static void rndis_filter_receive_response(struct rndis_device *dev,
+                                      struct rndis_message *resp)
+{
+       struct rndis_request *request = NULL;
+       bool found = false;
+       unsigned long flags;
+       struct net_device *ndev;
+
+       ndev = dev->net_dev->ndev;
+
+       spin_lock_irqsave(&dev->request_lock, flags);
+       list_for_each_entry(request, &dev->req_list, list_ent) {
+               /*
+                * All request/response message contains RequestId as the 1st
+                * field
+                */
+               if (request->request_msg.msg.init_req.req_id
+                   == resp->msg.init_complete.req_id) {
+                       found = true;
+                       break;
+               }
+       }
+       spin_unlock_irqrestore(&dev->request_lock, flags);
+
+       if (found) {
+               if (resp->msg_len <=
+                   sizeof(struct rndis_message) + RNDIS_EXT_LEN) {
+                       memcpy(&request->response_msg, resp,
+                              resp->msg_len);
+                       if (request->request_msg.ndis_msg_type ==
+                           RNDIS_MSG_QUERY && request->request_msg.msg.
+                           query_req.oid == RNDIS_OID_GEN_MEDIA_CONNECT_STATUS)
+                               rndis_set_link_state(dev, request);
+               } else {
+                       netdev_err(ndev,
+                               "rndis response buffer overflow "
+                               "detected (size %u max %zu)\n",
+                               resp->msg_len,
+                               sizeof(struct rndis_message));
+
+                       if (resp->ndis_msg_type ==
+                           RNDIS_MSG_RESET_C) {
+                               /* does not have a request id field */
+                               request->response_msg.msg.reset_complete.
+                                       status = RNDIS_STATUS_BUFFER_OVERFLOW;
+                       } else {
+                               request->response_msg.msg.
+                               init_complete.status =
+                                       RNDIS_STATUS_BUFFER_OVERFLOW;
+                       }
+               }
+
+               complete(&request->wait_event);
+       } else {
+               netdev_err(ndev,
+                       "no rndis request found for this response "
+                       "(id 0x%x res type 0x%x)\n",
+                       resp->msg.init_complete.req_id,
+                       resp->ndis_msg_type);
+       }
+}
+
+/*
+ * Get the Per-Packet-Info with the specified type
+ * return NULL if not found.
+ */
+static inline void *rndis_get_ppi(struct rndis_packet *rpkt, u32 type)
+{
+       struct rndis_per_packet_info *ppi;
+       int len;
+
+       if (rpkt->per_pkt_info_offset == 0)
+               return NULL;
+
+       ppi = (struct rndis_per_packet_info *)((ulong)rpkt +
+               rpkt->per_pkt_info_offset);
+       len = rpkt->per_pkt_info_len;
+
+       while (len > 0) {
+               if (ppi->type == type)
+                       return (void *)((ulong)ppi + ppi->ppi_offset);
+               len -= ppi->size;
+               ppi = (struct rndis_per_packet_info *)((ulong)ppi + ppi->size);
+       }
+
+       return NULL;
+}
+
+static void rndis_filter_receive_data(struct rndis_device *dev,
+                                  struct rndis_message *msg,
+                                  struct hv_netvsc_packet *pkt)
+{
+       struct rndis_packet *rndis_pkt;
+       u32 data_offset;
+       struct ndis_pkt_8021q_info *vlan;
+       struct ndis_tcp_ip_checksum_info *csum_info;
+
+       rndis_pkt = &msg->msg.pkt;
+
+       /* Remove the rndis header and pass it back up the stack */
+       data_offset = RNDIS_HEADER_SIZE + rndis_pkt->data_offset;
+
+       pkt->total_data_buflen -= data_offset;
+
+       /*
+        * Make sure we got a valid RNDIS message, now total_data_buflen
+        * should be the data packet size plus the trailer padding size
+        */
+       if (pkt->total_data_buflen < rndis_pkt->data_len) {
+               netdev_err(dev->net_dev->ndev, "rndis message buffer "
+                          "overflow detected (got %u, min %u)"
+                          "...dropping this message!\n",
+                          pkt->total_data_buflen, rndis_pkt->data_len);
+               return;
+       }
+
+       /*
+        * Remove the rndis trailer padding from rndis packet message
+        * rndis_pkt->data_len tell us the real data length, we only copy
+        * the data packet to the stack, without the rndis trailer padding
+        */
+       pkt->total_data_buflen = rndis_pkt->data_len;
+       pkt->data = (void *)((unsigned long)pkt->data + data_offset);
+
+       vlan = rndis_get_ppi(rndis_pkt, IEEE_8021Q_INFO);
+       if (vlan) {
+               pkt->vlan_tci = VLAN_TAG_PRESENT | vlan->vlanid |
+                       (vlan->pri << VLAN_PRIO_SHIFT);
+       } else {
+               pkt->vlan_tci = 0;
+       }
+
+       csum_info = rndis_get_ppi(rndis_pkt, TCPIP_CHKSUM_PKTINFO);
+       netvsc_recv_callback(dev->net_dev->dev, pkt, csum_info);
+}
+
+int rndis_filter_receive(struct hv_device *dev,
+                               struct hv_netvsc_packet *pkt)
+{
+       struct netvsc_device *net_dev = hv_get_drvdata(dev);
+       struct rndis_device *rndis_dev;
+       struct rndis_message *rndis_msg;
+       struct net_device *ndev;
+       int ret = 0;
+
+       if (!net_dev) {
+               ret = -EINVAL;
+               goto exit;
+       }
+
+       ndev = net_dev->ndev;
+
+       /* Make sure the rndis device state is initialized */
+       if (!net_dev->extension) {
+               netdev_err(ndev, "got rndis message but no rndis device - "
+                         "dropping this message!\n");
+               ret = -ENODEV;
+               goto exit;
+       }
+
+       rndis_dev = (struct rndis_device *)net_dev->extension;
+       if (rndis_dev->state == RNDIS_DEV_UNINITIALIZED) {
+               netdev_err(ndev, "got rndis message but rndis device "
+                          "uninitialized...dropping this message!\n");
+               ret = -ENODEV;
+               goto exit;
+       }
+
+       rndis_msg = pkt->data;
+
+       if (netif_msg_rx_err(net_dev->nd_ctx))
+               dump_rndis_message(dev, rndis_msg);
+
+       switch (rndis_msg->ndis_msg_type) {
+       case RNDIS_MSG_PACKET:
+               /* data msg */
+               rndis_filter_receive_data(rndis_dev, rndis_msg, pkt);
+               break;
+
+       case RNDIS_MSG_INIT_C:
+       case RNDIS_MSG_QUERY_C:
+       case RNDIS_MSG_SET_C:
+               /* completion msgs */
+               rndis_filter_receive_response(rndis_dev, rndis_msg);
+               break;
+
+       case RNDIS_MSG_INDICATE:
+               /* notification msgs */
+               netvsc_linkstatus_callback(dev, rndis_msg);
+               break;
+       default:
+               netdev_err(ndev,
+                       "unhandled rndis message (type %u len %u)\n",
+                          rndis_msg->ndis_msg_type,
+                          rndis_msg->msg_len);
+               break;
+       }
+
+exit:
+       if (ret != 0)
+               pkt->status = NVSP_STAT_FAIL;
+
+       return ret;
+}
+
+static int rndis_filter_query_device(struct rndis_device *dev, u32 oid,
+                                 void *result, u32 *result_size)
+{
+       struct rndis_request *request;
+       u32 inresult_size = *result_size;
+       struct rndis_query_request *query;
+       struct rndis_query_complete *query_complete;
+       int ret = 0;
+       unsigned long t;
+
+       if (!result)
+               return -EINVAL;
+
+       *result_size = 0;
+       request = get_rndis_request(dev, RNDIS_MSG_QUERY,
+                       RNDIS_MESSAGE_SIZE(struct rndis_query_request));
+       if (!request) {
+               ret = -ENOMEM;
+               goto cleanup;
+       }
+
+       /* Setup the rndis query */
+       query = &request->request_msg.msg.query_req;
+       query->oid = oid;
+       query->info_buf_offset = sizeof(struct rndis_query_request);
+       query->info_buflen = 0;
+       query->dev_vc_handle = 0;
+
+       if (oid == OID_GEN_RECEIVE_SCALE_CAPABILITIES) {
+               struct ndis_recv_scale_cap *cap;
+
+               request->request_msg.msg_len +=
+                       sizeof(struct ndis_recv_scale_cap);
+               query->info_buflen = sizeof(struct ndis_recv_scale_cap);
+               cap = (struct ndis_recv_scale_cap *)((unsigned long)query +
+                                                    query->info_buf_offset);
+               cap->hdr.type = NDIS_OBJECT_TYPE_RSS_CAPABILITIES;
+               cap->hdr.rev = NDIS_RECEIVE_SCALE_CAPABILITIES_REVISION_2;
+               cap->hdr.size = sizeof(struct ndis_recv_scale_cap);
+       }
+
+       ret = rndis_filter_send_request(dev, request);
+       if (ret != 0)
+               goto cleanup;
+
+       t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
+       if (t == 0) {
+               ret = -ETIMEDOUT;
+               goto cleanup;
+       }
+
+       /* Copy the response back */
+       query_complete = &request->response_msg.msg.query_complete;
+
+       if (query_complete->info_buflen > inresult_size) {
+               ret = -1;
+               goto cleanup;
+       }
+
+       memcpy(result,
+              (void *)((unsigned long)query_complete +
+                        query_complete->info_buf_offset),
+              query_complete->info_buflen);
+
+       *result_size = query_complete->info_buflen;
+
+cleanup:
+       if (request)
+               put_rndis_request(dev, request);
+
+       return ret;
+}
+
+static int rndis_filter_query_device_mac(struct rndis_device *dev)
+{
+       u32 size = ETH_ALEN;
+
+       return rndis_filter_query_device(dev,
+                                     RNDIS_OID_802_3_PERMANENT_ADDRESS,
+                                     dev->hw_mac_adr, &size);
+}
+
+#define NWADR_STR "NetworkAddress"
+#define NWADR_STRLEN 14
+
+int rndis_filter_set_device_mac(struct hv_device *hdev, char *mac)
+{
+       struct netvsc_device *nvdev = hv_get_drvdata(hdev);
+       struct rndis_device *rdev = nvdev->extension;
+       struct net_device *ndev = nvdev->ndev;
+       struct rndis_request *request;
+       struct rndis_set_request *set;
+       struct rndis_config_parameter_info *cpi;
+       wchar_t *cfg_nwadr, *cfg_mac;
+       struct rndis_set_complete *set_complete;
+       char macstr[2*ETH_ALEN+1];
+       u32 extlen = sizeof(struct rndis_config_parameter_info) +
+               2*NWADR_STRLEN + 4*ETH_ALEN;
+       int ret;
+       unsigned long t;
+
+       request = get_rndis_request(rdev, RNDIS_MSG_SET,
+               RNDIS_MESSAGE_SIZE(struct rndis_set_request) + extlen);
+       if (!request)
+               return -ENOMEM;
+
+       set = &request->request_msg.msg.set_req;
+       set->oid = RNDIS_OID_GEN_RNDIS_CONFIG_PARAMETER;
+       set->info_buflen = extlen;
+       set->info_buf_offset = sizeof(struct rndis_set_request);
+       set->dev_vc_handle = 0;
+
+       cpi = (struct rndis_config_parameter_info *)((ulong)set +
+               set->info_buf_offset);
+       cpi->parameter_name_offset =
+               sizeof(struct rndis_config_parameter_info);
+       /* Multiply by 2 because host needs 2 bytes (utf16) for each char */
+       cpi->parameter_name_length = 2*NWADR_STRLEN;
+       cpi->parameter_type = RNDIS_CONFIG_PARAM_TYPE_STRING;
+       cpi->parameter_value_offset =
+               cpi->parameter_name_offset + cpi->parameter_name_length;
+       /* Multiply by 4 because each MAC byte displayed as 2 utf16 chars */
+       cpi->parameter_value_length = 4*ETH_ALEN;
+
+       cfg_nwadr = (wchar_t *)((ulong)cpi + cpi->parameter_name_offset);
+       cfg_mac = (wchar_t *)((ulong)cpi + cpi->parameter_value_offset);
+       ret = utf8s_to_utf16s(NWADR_STR, NWADR_STRLEN, UTF16_HOST_ENDIAN,
+                             cfg_nwadr, NWADR_STRLEN);
+       if (ret < 0)
+               goto cleanup;
+       snprintf(macstr, 2*ETH_ALEN+1, "%pm", mac);
+       ret = utf8s_to_utf16s(macstr, 2*ETH_ALEN, UTF16_HOST_ENDIAN,
+                             cfg_mac, 2*ETH_ALEN);
+       if (ret < 0)
+               goto cleanup;
+
+       ret = rndis_filter_send_request(rdev, request);
+       if (ret != 0)
+               goto cleanup;
+
+       t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
+       if (t == 0) {
+               netdev_err(ndev, "timeout before we got a set response...\n");
+               /*
+                * can't put_rndis_request, since we may still receive a
+                * send-completion.
+                */
+               return -EBUSY;
+       } else {
+               set_complete = &request->response_msg.msg.set_complete;
+               if (set_complete->status != RNDIS_STATUS_SUCCESS) {
+                       netdev_err(ndev, "Fail to set MAC on host side:0x%x\n",
+                                  set_complete->status);
+                       ret = -EINVAL;
+               }
+       }
+
+cleanup:
+       put_rndis_request(rdev, request);
+       return ret;
+}
+
+static int
+rndis_filter_set_offload_params(struct hv_device *hdev,
+                               struct ndis_offload_params *req_offloads)
+{
+       struct netvsc_device *nvdev = hv_get_drvdata(hdev);
+       struct rndis_device *rdev = nvdev->extension;
+       struct net_device *ndev = nvdev->ndev;
+       struct rndis_request *request;
+       struct rndis_set_request *set;
+       struct ndis_offload_params *offload_params;
+       struct rndis_set_complete *set_complete;
+       u32 extlen = sizeof(struct ndis_offload_params);
+       int ret;
+       unsigned long t;
+       u32 vsp_version = nvdev->nvsp_version;
+
+       if (vsp_version <= NVSP_PROTOCOL_VERSION_4) {
+               extlen = VERSION_4_OFFLOAD_SIZE;
+               /* On NVSP_PROTOCOL_VERSION_4 and below, we do not support
+                * UDP checksum offload.
+                */
+               req_offloads->udp_ip_v4_csum = 0;
+               req_offloads->udp_ip_v6_csum = 0;
+       }
+
+       request = get_rndis_request(rdev, RNDIS_MSG_SET,
+               RNDIS_MESSAGE_SIZE(struct rndis_set_request) + extlen);
+       if (!request)
+               return -ENOMEM;
+
+       set = &request->request_msg.msg.set_req;
+       set->oid = OID_TCP_OFFLOAD_PARAMETERS;
+       set->info_buflen = extlen;
+       set->info_buf_offset = sizeof(struct rndis_set_request);
+       set->dev_vc_handle = 0;
+
+       offload_params = (struct ndis_offload_params *)((ulong)set +
+                               set->info_buf_offset);
+       *offload_params = *req_offloads;
+       offload_params->header.type = NDIS_OBJECT_TYPE_DEFAULT;
+       offload_params->header.revision = NDIS_OFFLOAD_PARAMETERS_REVISION_3;
+       offload_params->header.size = extlen;
+
+       ret = rndis_filter_send_request(rdev, request);
+       if (ret != 0)
+               goto cleanup;
+
+       t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
+       if (t == 0) {
+               netdev_err(ndev, "timeout before we got aOFFLOAD set response...\n");
+               /* can't put_rndis_request, since we may still receive a
+                * send-completion.
+                */
+               return -EBUSY;
+       } else {
+               set_complete = &request->response_msg.msg.set_complete;
+               if (set_complete->status != RNDIS_STATUS_SUCCESS) {
+                       netdev_err(ndev, "Fail to set offload on host side:0x%x\n",
+                                  set_complete->status);
+                       ret = -EINVAL;
+               }
+       }
+
+cleanup:
+       put_rndis_request(rdev, request);
+       return ret;
+}
+
+u8 netvsc_hash_key[HASH_KEYLEN] = {
+       0x6d, 0x5a, 0x56, 0xda, 0x25, 0x5b, 0x0e, 0xc2,
+       0x41, 0x67, 0x25, 0x3d, 0x43, 0xa3, 0x8f, 0xb0,
+       0xd0, 0xca, 0x2b, 0xcb, 0xae, 0x7b, 0x30, 0xb4,
+       0x77, 0xcb, 0x2d, 0xa3, 0x80, 0x30, 0xf2, 0x0c,
+       0x6a, 0x42, 0xb7, 0x3b, 0xbe, 0xac, 0x01, 0xfa
+};
+
+static int rndis_filter_set_rss_param(struct rndis_device *rdev, int num_queue)
+{
+       struct net_device *ndev = rdev->net_dev->ndev;
+       struct rndis_request *request;
+       struct rndis_set_request *set;
+       struct rndis_set_complete *set_complete;
+       u32 extlen = sizeof(struct ndis_recv_scale_param) +
+                    4*ITAB_NUM + HASH_KEYLEN;
+       struct ndis_recv_scale_param *rssp;
+       u32 *itab;
+       u8 *keyp;
+       int i, ret;
+       unsigned long t;
+
+       request = get_rndis_request(
+                       rdev, RNDIS_MSG_SET,
+                       RNDIS_MESSAGE_SIZE(struct rndis_set_request) + extlen);
+       if (!request)
+               return -ENOMEM;
+
+       set = &request->request_msg.msg.set_req;
+       set->oid = OID_GEN_RECEIVE_SCALE_PARAMETERS;
+       set->info_buflen = extlen;
+       set->info_buf_offset = sizeof(struct rndis_set_request);
+       set->dev_vc_handle = 0;
+
+       rssp = (struct ndis_recv_scale_param *)(set + 1);
+       rssp->hdr.type = NDIS_OBJECT_TYPE_RSS_PARAMETERS;
+       rssp->hdr.rev = NDIS_RECEIVE_SCALE_PARAMETERS_REVISION_2;
+       rssp->hdr.size = sizeof(struct ndis_recv_scale_param);
+       rssp->flag = 0;
+       rssp->hashinfo = NDIS_HASH_FUNC_TOEPLITZ | NDIS_HASH_IPV4 |
+                        NDIS_HASH_TCP_IPV4 | NDIS_HASH_IPV6 |
+                        NDIS_HASH_TCP_IPV6;
+       rssp->indirect_tabsize = 4*ITAB_NUM;
+       rssp->indirect_taboffset = sizeof(struct ndis_recv_scale_param);
+       rssp->hashkey_size = HASH_KEYLEN;
+       rssp->kashkey_offset = rssp->indirect_taboffset +
+                              rssp->indirect_tabsize;
+
+       /* Set indirection table entries */
+       itab = (u32 *)(rssp + 1);
+       for (i = 0; i < ITAB_NUM; i++)
+               itab[i] = i % num_queue;
+
+       /* Set hask key values */
+       keyp = (u8 *)((unsigned long)rssp + rssp->kashkey_offset);
+       for (i = 0; i < HASH_KEYLEN; i++)
+               keyp[i] = netvsc_hash_key[i];
+
+
+       ret = rndis_filter_send_request(rdev, request);
+       if (ret != 0)
+               goto cleanup;
+
+       t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
+       if (t == 0) {
+               netdev_err(ndev, "timeout before we got a set response...\n");
+               /* can't put_rndis_request, since we may still receive a
+                * send-completion.
+                */
+               return -ETIMEDOUT;
+       } else {
+               set_complete = &request->response_msg.msg.set_complete;
+               if (set_complete->status != RNDIS_STATUS_SUCCESS) {
+                       netdev_err(ndev, "Fail to set RSS parameters:0x%x\n",
+                                  set_complete->status);
+                       ret = -EINVAL;
+               }
+       }
+
+cleanup:
+       put_rndis_request(rdev, request);
+       return ret;
+}
+
+
+static int rndis_filter_query_device_link_status(struct rndis_device *dev)
+{
+       u32 size = sizeof(u32);
+       u32 link_status;
+       int ret;
+
+       ret = rndis_filter_query_device(dev,
+                                     RNDIS_OID_GEN_MEDIA_CONNECT_STATUS,
+                                     &link_status, &size);
+
+       return ret;
+}
+
+int rndis_filter_set_packet_filter(struct rndis_device *dev, u32 new_filter)
+{
+       struct rndis_request *request;
+       struct rndis_set_request *set;
+       struct rndis_set_complete *set_complete;
+       u32 status;
+       int ret;
+       unsigned long t;
+       struct net_device *ndev;
+
+       ndev = dev->net_dev->ndev;
+
+       request = get_rndis_request(dev, RNDIS_MSG_SET,
+                       RNDIS_MESSAGE_SIZE(struct rndis_set_request) +
+                       sizeof(u32));
+       if (!request) {
+               ret = -ENOMEM;
+               goto cleanup;
+       }
+
+       /* Setup the rndis set */
+       set = &request->request_msg.msg.set_req;
+       set->oid = RNDIS_OID_GEN_CURRENT_PACKET_FILTER;
+       set->info_buflen = sizeof(u32);
+       set->info_buf_offset = sizeof(struct rndis_set_request);
+
+       memcpy((void *)(unsigned long)set + sizeof(struct rndis_set_request),
+              &new_filter, sizeof(u32));
+
+       ret = rndis_filter_send_request(dev, request);
+       if (ret != 0)
+               goto cleanup;
+
+       t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
+
+       if (t == 0) {
+               netdev_err(ndev,
+                       "timeout before we got a set response...\n");
+               ret = -ETIMEDOUT;
+               /*
+                * We can't deallocate the request since we may still receive a
+                * send completion for it.
+                */
+               goto exit;
+       } else {
+               set_complete = &request->response_msg.msg.set_complete;
+               status = set_complete->status;
+       }
+
+cleanup:
+       if (request)
+               put_rndis_request(dev, request);
+exit:
+       return ret;
+}
+
+
+static int rndis_filter_init_device(struct rndis_device *dev)
+{
+       struct rndis_request *request;
+       struct rndis_initialize_request *init;
+       struct rndis_initialize_complete *init_complete;
+       u32 status;
+       int ret;
+       unsigned long t;
+       struct netvsc_device *nvdev = dev->net_dev;
+
+       request = get_rndis_request(dev, RNDIS_MSG_INIT,
+                       RNDIS_MESSAGE_SIZE(struct rndis_initialize_request));
+       if (!request) {
+               ret = -ENOMEM;
+               goto cleanup;
+       }
+
+       /* Setup the rndis set */
+       init = &request->request_msg.msg.init_req;
+       init->major_ver = RNDIS_MAJOR_VERSION;
+       init->minor_ver = RNDIS_MINOR_VERSION;
+       init->max_xfer_size = 0x4000;
+
+       dev->state = RNDIS_DEV_INITIALIZING;
+
+       ret = rndis_filter_send_request(dev, request);
+       if (ret != 0) {
+               dev->state = RNDIS_DEV_UNINITIALIZED;
+               goto cleanup;
+       }
+
+
+       t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
+
+       if (t == 0) {
+               ret = -ETIMEDOUT;
+               goto cleanup;
+       }
+
+       init_complete = &request->response_msg.msg.init_complete;
+       status = init_complete->status;
+       if (status == RNDIS_STATUS_SUCCESS) {
+               dev->state = RNDIS_DEV_INITIALIZED;
+               nvdev->max_pkt = init_complete->max_pkt_per_msg;
+               nvdev->pkt_align = 1 << init_complete->pkt_alignment_factor;
+               ret = 0;
+       } else {
+               dev->state = RNDIS_DEV_UNINITIALIZED;
+               ret = -EINVAL;
+       }
+
+cleanup:
+       if (request)
+               put_rndis_request(dev, request);
+
+       return ret;
+}
+
+static void rndis_filter_halt_device(struct rndis_device *dev)
+{
+       struct rndis_request *request;
+       struct rndis_halt_request *halt;
+       struct netvsc_device *nvdev = dev->net_dev;
+       struct hv_device *hdev = nvdev->dev;
+       ulong flags;
+
+       /* Attempt to do a rndis device halt */
+       request = get_rndis_request(dev, RNDIS_MSG_HALT,
+                               RNDIS_MESSAGE_SIZE(struct rndis_halt_request));
+       if (!request)
+               goto cleanup;
+
+       /* Setup the rndis set */
+       halt = &request->request_msg.msg.halt_req;
+       halt->req_id = atomic_inc_return(&dev->new_req_id);
+
+       /* Ignore return since this msg is optional. */
+       rndis_filter_send_request(dev, request);
+
+       dev->state = RNDIS_DEV_UNINITIALIZED;
+
+cleanup:
+       spin_lock_irqsave(&hdev->channel->inbound_lock, flags);
+       nvdev->destroy = true;
+       spin_unlock_irqrestore(&hdev->channel->inbound_lock, flags);
+
+       /* Wait for all send completions */
+       wait_event(nvdev->wait_drain,
+               atomic_read(&nvdev->num_outstanding_sends) == 0);
+
+       if (request)
+               put_rndis_request(dev, request);
+       return;
+}
+
+static int rndis_filter_open_device(struct rndis_device *dev)
+{
+       int ret;
+
+       if (dev->state != RNDIS_DEV_INITIALIZED)
+               return 0;
+
+       ret = rndis_filter_set_packet_filter(dev,
+                                        NDIS_PACKET_TYPE_BROADCAST |
+                                        NDIS_PACKET_TYPE_ALL_MULTICAST |
+                                        NDIS_PACKET_TYPE_DIRECTED);
+       if (ret == 0)
+               dev->state = RNDIS_DEV_DATAINITIALIZED;
+
+       return ret;
+}
+
+static int rndis_filter_close_device(struct rndis_device *dev)
+{
+       int ret;
+
+       if (dev->state != RNDIS_DEV_DATAINITIALIZED)
+               return 0;
+
+       ret = rndis_filter_set_packet_filter(dev, 0);
+       if (ret == -ENODEV)
+               ret = 0;
+
+       if (ret == 0)
+               dev->state = RNDIS_DEV_INITIALIZED;
+
+       return ret;
+}
+
+static void netvsc_sc_open(struct vmbus_channel *new_sc)
+{
+       struct netvsc_device *nvscdev;
+       u16 chn_index = new_sc->offermsg.offer.sub_channel_index;
+       int ret;
+
+       nvscdev = hv_get_drvdata(new_sc->primary_channel->device_obj);
+
+       if (chn_index >= nvscdev->num_chn)
+               return;
+
+       set_per_channel_state(new_sc, nvscdev->sub_cb_buf + (chn_index - 1) *
+                             NETVSC_PACKET_SIZE);
+
+       ret = vmbus_open(new_sc, nvscdev->ring_size * PAGE_SIZE,
+                        nvscdev->ring_size * PAGE_SIZE, NULL, 0,
+                        netvsc_channel_cb, new_sc);
+
+       if (ret == 0)
+               nvscdev->chn_table[chn_index] = new_sc;
+}
+
+int rndis_filter_device_add(struct hv_device *dev,
+                                 void *additional_info)
+{
+       int ret;
+       struct netvsc_device *net_device;
+       struct rndis_device *rndis_device;
+       struct netvsc_device_info *device_info = additional_info;
+       struct ndis_offload_params offloads;
+       struct nvsp_message *init_packet;
+       unsigned long t;
+       struct ndis_recv_scale_cap rsscap;
+       u32 rsscap_size = sizeof(struct ndis_recv_scale_cap);
+       u32 mtu, size;
+
+       rndis_device = get_rndis_device();
+       if (!rndis_device)
+               return -ENODEV;
+
+       /*
+        * Let the inner driver handle this first to create the netvsc channel
+        * NOTE! Once the channel is created, we may get a receive callback
+        * (RndisFilterOnReceive()) before this call is completed
+        */
+       ret = netvsc_device_add(dev, additional_info);
+       if (ret != 0) {
+               kfree(rndis_device);
+               return ret;
+       }
+
+
+       /* Initialize the rndis device */
+       net_device = hv_get_drvdata(dev);
+       net_device->max_chn = 1;
+       net_device->num_chn = 1;
+
+       net_device->extension = rndis_device;
+       rndis_device->net_dev = net_device;
+
+       /* Send the rndis initialization message */
+       ret = rndis_filter_init_device(rndis_device);
+       if (ret != 0) {
+               rndis_filter_device_remove(dev);
+               return ret;
+       }
+
+       /* Get the MTU from the host */
+       size = sizeof(u32);
+       ret = rndis_filter_query_device(rndis_device,
+                                       RNDIS_OID_GEN_MAXIMUM_FRAME_SIZE,
+                                       &mtu, &size);
+       if (ret == 0 && size == sizeof(u32))
+               net_device->ndev->mtu = mtu;
+
+       /* Get the mac address */
+       ret = rndis_filter_query_device_mac(rndis_device);
+       if (ret != 0) {
+               rndis_filter_device_remove(dev);
+               return ret;
+       }
+
+       memcpy(device_info->mac_adr, rndis_device->hw_mac_adr, ETH_ALEN);
+
+       /* Turn on the offloads; the host supports all of the relevant
+        * offloads.
+        */
+       memset(&offloads, 0, sizeof(struct ndis_offload_params));
+       /* A value of zero means "no change"; now turn on what we
+        * want.
+        */
+       offloads.ip_v4_csum = NDIS_OFFLOAD_PARAMETERS_TX_RX_ENABLED;
+       offloads.tcp_ip_v4_csum = NDIS_OFFLOAD_PARAMETERS_TX_RX_ENABLED;
+       offloads.udp_ip_v4_csum = NDIS_OFFLOAD_PARAMETERS_TX_RX_ENABLED;
+       offloads.tcp_ip_v6_csum = NDIS_OFFLOAD_PARAMETERS_TX_RX_ENABLED;
+       offloads.udp_ip_v6_csum = NDIS_OFFLOAD_PARAMETERS_TX_RX_ENABLED;
+       offloads.lso_v2_ipv4 = NDIS_OFFLOAD_PARAMETERS_LSOV2_ENABLED;
+
+
+       ret = rndis_filter_set_offload_params(dev, &offloads);
+       if (ret)
+               goto err_dev_remv;
+
+       rndis_filter_query_device_link_status(rndis_device);
+
+       device_info->link_state = rndis_device->link_state;
+
+       dev_info(&dev->device, "Device MAC %pM link state %s\n",
+                rndis_device->hw_mac_adr,
+                device_info->link_state ? "down" : "up");
+
+       if (net_device->nvsp_version < NVSP_PROTOCOL_VERSION_5)
+               return 0;
+
+       /* vRSS setup */
+       memset(&rsscap, 0, rsscap_size);
+       ret = rndis_filter_query_device(rndis_device,
+                                       OID_GEN_RECEIVE_SCALE_CAPABILITIES,
+                                       &rsscap, &rsscap_size);
+       if (ret || rsscap.num_recv_que < 2)
+               goto out;
+
+       net_device->max_chn = rsscap.num_recv_que;
+       net_device->num_chn = (num_online_cpus() < rsscap.num_recv_que) ?
+                              num_online_cpus() : rsscap.num_recv_que;
+       if (net_device->num_chn == 1)
+               goto out;
+
+       net_device->sub_cb_buf = vzalloc((net_device->num_chn - 1) *
+                                        NETVSC_PACKET_SIZE);
+       if (!net_device->sub_cb_buf) {
+               net_device->num_chn = 1;
+               dev_info(&dev->device, "No memory for subchannels.\n");
+               goto out;
+       }
+
+       vmbus_set_sc_create_callback(dev->channel, netvsc_sc_open);
+
+       init_packet = &net_device->channel_init_pkt;
+       memset(init_packet, 0, sizeof(struct nvsp_message));
+       init_packet->hdr.msg_type = NVSP_MSG5_TYPE_SUBCHANNEL;
+       init_packet->msg.v5_msg.subchn_req.op = NVSP_SUBCHANNEL_ALLOCATE;
+       init_packet->msg.v5_msg.subchn_req.num_subchannels =
+                                               net_device->num_chn - 1;
+       ret = vmbus_sendpacket(dev->channel, init_packet,
+                              sizeof(struct nvsp_message),
+                              (unsigned long)init_packet,
+                              VM_PKT_DATA_INBAND,
+                              VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
+       if (ret)
+               goto out;
+       t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
+       if (t == 0) {
+               ret = -ETIMEDOUT;
+               goto out;
+       }
+       if (init_packet->msg.v5_msg.subchn_comp.status !=
+           NVSP_STAT_SUCCESS) {
+               ret = -ENODEV;
+               goto out;
+       }
+       net_device->num_chn = 1 +
+               init_packet->msg.v5_msg.subchn_comp.num_subchannels;
+
+       ret = rndis_filter_set_rss_param(rndis_device, net_device->num_chn);
+
+out:
+       if (ret) {
+               net_device->max_chn = 1;
+               net_device->num_chn = 1;
+       }
+       return 0; /* return 0 because primary channel can be used alone */
+
+err_dev_remv:
+       rndis_filter_device_remove(dev);
+       return ret;
+}
+
+void rndis_filter_device_remove(struct hv_device *dev)
+{
+       struct netvsc_device *net_dev = hv_get_drvdata(dev);
+       struct rndis_device *rndis_dev = net_dev->extension;
+
+       /* Halt and release the rndis device */
+       rndis_filter_halt_device(rndis_dev);
+
+       kfree(rndis_dev);
+       net_dev->extension = NULL;
+
+       netvsc_device_remove(dev);
+}
+
+
+int rndis_filter_open(struct hv_device *dev)
+{
+       struct netvsc_device *net_device = hv_get_drvdata(dev);
+
+       if (!net_device)
+               return -EINVAL;
+
+       return rndis_filter_open_device(net_device->extension);
+}
+
+int rndis_filter_close(struct hv_device *dev)
+{
+       struct netvsc_device *nvdev = hv_get_drvdata(dev);
+
+       if (!nvdev)
+               return -EINVAL;
+
+       return rndis_filter_close_device(nvdev->extension);
+}