datapath-windows/ovsext/Gre.c

   1 /*
   2  * Copyright (c) 2015, 2016 Cloudbase Solutions Srl
   3  *
   4  * Licensed under the Apache License, Version 2.0 (the "License");
   5  * you may not use this file except in compliance with the License.
   6  * You may obtain a copy of the License at
   7  *
   8  *     http://www.apache.org/licenses/LICENSE-2.0
   9  *
  10  * Unless required by applicable law or agreed to in writing, software
  11  * distributed under the License is distributed on an "AS IS" BASIS,
  12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13  * See the License for the specific language governing permissions and
  14  * limitations under the License.
  15  */
  16
  17 #include "precomp.h"
  18
  19 #include "Atomic.h"
  20 #include "Debug.h"
  21 #include "Flow.h"
  22 #include "Gre.h"
  23 #include "IpHelper.h"
  24 #include "NetProto.h"
  25 #include "Offload.h"
  26 #include "PacketIO.h"
  27 #include "PacketParser.h"
  28 #include "Switch.h"
  29 #include "User.h"
  30 #include "Util.h"
  31 #include "Vport.h"
  32
  33 #ifdef OVS_DBG_MOD
  34 #undef OVS_DBG_MOD
  35 #endif
  36 #define OVS_DBG_MOD OVS_DBG_GRE
  37
  38 static NDIS_STATUS
  39 OvsDoEncapGre(POVS_VPORT_ENTRY vport, PNET_BUFFER_LIST curNbl,
  40               const OvsIPv4TunnelKey *tunKey,
  41               const POVS_FWD_INFO fwdInfo,
  42               POVS_PACKET_HDR_INFO layers,
  43               POVS_SWITCH_CONTEXT switchContext,
  44               PNET_BUFFER_LIST *newNbl);
  45
  46 /*
  47  * --------------------------------------------------------------------------
  48  * OvsInitGreTunnel --
  49  *    Initialize GRE tunnel module.
  50  * --------------------------------------------------------------------------
  51  */
  52 NTSTATUS
  53 OvsInitGreTunnel(POVS_VPORT_ENTRY vport)
  54 {
  55     POVS_GRE_VPORT grePort;
  56
  57     grePort = (POVS_GRE_VPORT)OvsAllocateMemoryWithTag(sizeof(*grePort),
  58                                                        OVS_GRE_POOL_TAG);
  59     if (!grePort) {
  60         OVS_LOG_ERROR("Insufficient memory, can't allocate OVS_GRE_VPORT");
  61         return STATUS_INSUFFICIENT_RESOURCES;
  62     }
  63
  64     RtlZeroMemory(grePort, sizeof(*grePort));
  65     vport->priv = (PVOID)grePort;
  66     return STATUS_SUCCESS;
  67 }
  68
  69 /*
  70  * --------------------------------------------------------------------------
  71  * OvsCleanupGreTunnel --
  72  *    Cleanup GRE Tunnel module.
  73  * --------------------------------------------------------------------------
  74  */
  75 void
  76 OvsCleanupGreTunnel(POVS_VPORT_ENTRY vport)
  77 {
  78     if (vport->ovsType != OVS_VPORT_TYPE_GRE ||
  79         vport->priv == NULL) {
  80         return;
  81     }
  82
  83     OvsFreeMemoryWithTag(vport->priv, OVS_GRE_POOL_TAG);
  84     vport->priv = NULL;
  85 }
  86
  87 /*
  88  * --------------------------------------------------------------------------
  89  * OvsEncapGre --
  90  *     Encapsulates a packet with an GRE header.
  91  * --------------------------------------------------------------------------
  92  */
  93 NDIS_STATUS
  94 OvsEncapGre(POVS_VPORT_ENTRY vport,
  95             PNET_BUFFER_LIST curNbl,
  96             OvsIPv4TunnelKey *tunKey,
  97             POVS_SWITCH_CONTEXT switchContext,
  98             POVS_PACKET_HDR_INFO layers,
  99             PNET_BUFFER_LIST *newNbl)
 100 {
 101     OVS_FWD_INFO fwdInfo;
 102     NDIS_STATUS status;
 103
 104     status = OvsLookupIPFwdInfo(tunKey->dst, &fwdInfo);
 105     if (status != STATUS_SUCCESS) {
 106         OvsFwdIPHelperRequest(NULL, 0, tunKey, NULL, NULL, NULL);
 107         return NDIS_STATUS_FAILURE;
 108     }
 109
 110     status = OvsDoEncapGre(vport, curNbl, tunKey, &fwdInfo, layers,
 111                            switchContext, newNbl);
 112     return status;
 113 }
 114
 115 /*
 116  * --------------------------------------------------------------------------
 117  * OvsDoEncapGre --
 118  *    Internal utility function which actually does the GRE encap.
 119  * --------------------------------------------------------------------------
 120  */
 121 NDIS_STATUS
 122 OvsDoEncapGre(POVS_VPORT_ENTRY vport,
 123               PNET_BUFFER_LIST curNbl,
 124               const OvsIPv4TunnelKey *tunKey,
 125               const POVS_FWD_INFO fwdInfo,
 126               POVS_PACKET_HDR_INFO layers,
 127               POVS_SWITCH_CONTEXT switchContext,
 128               PNET_BUFFER_LIST *newNbl)
 129 {
 130     NDIS_STATUS status;
 131     PNET_BUFFER curNb;
 132     PMDL curMdl;
 133     PUINT8 bufferStart;
 134     EthHdr *ethHdr;
 135     IPHdr *ipHdr;
 136     PGREHdr greHdr;
 137     POVS_GRE_VPORT vportGre;
 138     UINT32 headRoom = GreTunHdrSize(tunKey->flags);
 139 #if DBG
 140     UINT32 counterHeadRoom;
 141 #endif
 142     UINT32 packetLength;
 143     ULONG mss = 0;
 144     ASSERT(*newNbl == NULL);
 145
 146     curNb = NET_BUFFER_LIST_FIRST_NB(curNbl);
 147     packetLength = NET_BUFFER_DATA_LENGTH(curNb);
 148
 149     if (layers->isTcp) {
 150         mss = OVSGetTcpMSS(curNbl);
 151
 152         OVS_LOG_TRACE("MSS %u packet len %u", mss,
 153                       packetLength);
 154         if (mss) {
 155             OVS_LOG_TRACE("l4Offset %d", layers->l4Offset);
 156             *newNbl = OvsTcpSegmentNBL(switchContext, curNbl, layers,
 157                                        mss, headRoom);
 158             if (*newNbl == NULL) {
 159                 OVS_LOG_ERROR("Unable to segment NBL");
 160                 return NDIS_STATUS_FAILURE;
 161             }
 162             /* Clear out LSO flags after this point */
 163             NET_BUFFER_LIST_INFO(*newNbl, TcpLargeSendNetBufferListInfo) = 0;
 164         }
 165     }
 166
 167     vportGre = (POVS_GRE_VPORT)GetOvsVportPriv(vport);
 168     ASSERT(vportGre);
 169
 170     /* If we didn't split the packet above, make a copy now */
 171     if (*newNbl == NULL) {
 172         *newNbl = OvsPartialCopyNBL(switchContext, curNbl, 0, headRoom,
 173                                     FALSE /*NBL info*/);
 174         if (*newNbl == NULL) {
 175             OVS_LOG_ERROR("Unable to copy NBL");
 176             return NDIS_STATUS_FAILURE;
 177         }
 178
 179         NDIS_TCP_IP_CHECKSUM_NET_BUFFER_LIST_INFO csumInfo;
 180         csumInfo.Value = NET_BUFFER_LIST_INFO(curNbl,
 181                                               TcpIpChecksumNetBufferListInfo);
 182
 183         status = OvsApplySWChecksumOnNB(layers, *newNbl, &csumInfo);
 184         if (status != NDIS_STATUS_SUCCESS) {
 185             goto ret_error;
 186         }
 187     }
 188
 189     curNbl = *newNbl;
 190     for (curNb = NET_BUFFER_LIST_FIRST_NB(curNbl); curNb != NULL;
 191          curNb = curNb->Next) {
 192 #if DBG
 193         counterHeadRoom = headRoom;
 194 #endif
 195         status = NdisRetreatNetBufferDataStart(curNb, headRoom, 0, NULL);
 196         if (status != NDIS_STATUS_SUCCESS) {
 197             goto ret_error;
 198         }
 199
 200         curMdl = NET_BUFFER_CURRENT_MDL(curNb);
 201         bufferStart = (PUINT8)MmGetSystemAddressForMdlSafe(curMdl,
 202                                                            LowPagePriority);
 203         if (!bufferStart) {
 204             status = NDIS_STATUS_RESOURCES;
 205             goto ret_error;
 206         }
 207
 208         bufferStart += NET_BUFFER_CURRENT_MDL_OFFSET(curNb);
 209         if (NET_BUFFER_NEXT_NB(curNb)) {
 210             OVS_LOG_TRACE("nb length %u next %u",
 211                           NET_BUFFER_DATA_LENGTH(curNb),
 212                           NET_BUFFER_DATA_LENGTH(curNb->Next));
 213         }
 214
 215         /* L2 header */
 216         ethHdr = (EthHdr *)bufferStart;
 217         ASSERT(((PCHAR)&fwdInfo->dstMacAddr + sizeof fwdInfo->dstMacAddr) ==
 218                (PCHAR)&fwdInfo->srcMacAddr);
 219         NdisMoveMemory(ethHdr->Destination, fwdInfo->dstMacAddr,
 220                        sizeof ethHdr->Destination + sizeof ethHdr->Source);
 221         ethHdr->Type = htons(ETH_TYPE_IPV4);
 222 #if DBG
 223         counterHeadRoom -= sizeof *ethHdr;
 224 #endif
 225
 226         /* IP header */
 227         ipHdr = (IPHdr *)((PCHAR)ethHdr + sizeof *ethHdr);
 228
 229         ipHdr->ihl = sizeof *ipHdr / 4;
 230         ipHdr->version = IPPROTO_IPV4;
 231         ipHdr->tos = tunKey->tos;
 232         ipHdr->tot_len = htons(NET_BUFFER_DATA_LENGTH(curNb) - sizeof *ethHdr);
 233         ipHdr->id = (uint16)atomic_add64(&vportGre->ipId,
 234                                          NET_BUFFER_DATA_LENGTH(curNb));
 235         ipHdr->frag_off = (tunKey->flags & OVS_TNL_F_DONT_FRAGMENT) ?
 236                           IP_DF_NBO : 0;
 237         ipHdr->ttl = tunKey->ttl ? tunKey->ttl : 64;
 238         ipHdr->protocol = IPPROTO_GRE;
 239         ASSERT(tunKey->dst == fwdInfo->dstIpAddr);
 240         ASSERT(tunKey->src == fwdInfo->srcIpAddr || tunKey->src == 0);
 241         ipHdr->saddr = fwdInfo->srcIpAddr;
 242         ipHdr->daddr = fwdInfo->dstIpAddr;
 243
 244         ipHdr->check = 0;
 245         ipHdr->check = IPChecksum((UINT8 *)ipHdr, sizeof *ipHdr, 0);
 246 #if DBG
 247         counterHeadRoom -= sizeof *ipHdr;
 248 #endif
 249
 250         /* GRE header */
 251         greHdr = (GREHdr *)((PCHAR)ipHdr + sizeof *ipHdr);
 252         greHdr->flags = OvsTunnelFlagsToGreFlags(tunKey->flags);
 253         greHdr->protocolType = GRE_NET_TEB;
 254 #if DBG
 255         counterHeadRoom -= sizeof *greHdr;
 256 #endif
 257
 258         PCHAR currentOffset = (PCHAR)greHdr + sizeof *greHdr;
 259
 260         if (tunKey->flags & OVS_TNL_F_CSUM) {
 261             RtlZeroMemory(currentOffset, 4);
 262             currentOffset += 4;
 263 #if DBG
 264             counterHeadRoom -= 4;
 265 #endif
 266         }
 267
 268         if (tunKey->flags & OVS_TNL_F_KEY) {
 269             RtlZeroMemory(currentOffset, 4);
 270             UINT32 key = (tunKey->tunnelId >> 32);
 271             RtlCopyMemory(currentOffset, &key, sizeof key);
 272             currentOffset += 4;
 273 #if DBG
 274             counterHeadRoom -= 4;
 275 #endif
 276         }
 277
 278 #if DBG
 279         ASSERT(counterHeadRoom == 0);
 280 #endif
 281
 282     }
 283     return STATUS_SUCCESS;
 284
 285 ret_error:
 286     OvsCompleteNBL(switchContext, *newNbl, TRUE);
 287     *newNbl = NULL;
 288     return status;
 289 }
 290
 291 NDIS_STATUS
 292 OvsDecapGre(POVS_SWITCH_CONTEXT switchContext,
 293             PNET_BUFFER_LIST curNbl,
 294             OvsIPv4TunnelKey *tunKey,
 295             PNET_BUFFER_LIST *newNbl)
 296 {
 297     PNET_BUFFER curNb;
 298     PMDL curMdl;
 299     EthHdr *ethHdr;
 300     IPHdr *ipHdr;
 301     GREHdr *greHdr;
 302     UINT32 tunnelSize = 0, packetLength = 0;
 303     UINT32 headRoom = 0;
 304     PUINT8 bufferStart;
 305     NDIS_STATUS status;
 306
 307     curNb = NET_BUFFER_LIST_FIRST_NB(curNbl);
 308     packetLength = NET_BUFFER_DATA_LENGTH(curNb);
 309     tunnelSize = GreTunHdrSize(tunKey->flags);
 310     if (packetLength <= tunnelSize) {
 311         return NDIS_STATUS_INVALID_LENGTH;
 312     }
 313
 314     /*
 315      * Create a copy of the NBL so that we have all the headers in one MDL.
 316      */
 317     *newNbl = OvsPartialCopyNBL(switchContext, curNbl,
 318                                 tunnelSize + OVS_DEFAULT_COPY_SIZE, 0,
 319                                 TRUE /*copy NBL info */);
 320
 321     if (*newNbl == NULL) {
 322         return NDIS_STATUS_RESOURCES;
 323     }
 324
 325     curNbl = *newNbl;
 326     curNb = NET_BUFFER_LIST_FIRST_NB(curNbl);
 327     curMdl = NET_BUFFER_CURRENT_MDL(curNb);
 328     bufferStart = (PUINT8)MmGetSystemAddressForMdlSafe(curMdl, LowPagePriority) +
 329                   NET_BUFFER_CURRENT_MDL_OFFSET(curNb);
 330     if (!bufferStart) {
 331         status = NDIS_STATUS_RESOURCES;
 332         goto dropNbl;
 333     }
 334
 335     ethHdr = (EthHdr *)bufferStart;
 336     headRoom += sizeof *ethHdr;
 337
 338     ipHdr = (IPHdr *)((PCHAR)ethHdr + sizeof *ethHdr);
 339     tunKey->src = ipHdr->saddr;
 340     tunKey->dst = ipHdr->daddr;
 341     tunKey->tos = ipHdr->tos;
 342     tunKey->ttl = ipHdr->ttl;
 343     tunKey->pad = 0;
 344     headRoom += sizeof *ipHdr;
 345
 346     greHdr = (GREHdr *)((PCHAR)ipHdr + sizeof *ipHdr);
 347     headRoom += sizeof *greHdr;
 348
 349     /* Validate if GRE header protocol type. */
 350     if (greHdr->protocolType != GRE_NET_TEB) {
 351         status = STATUS_NDIS_INVALID_PACKET;
 352         goto dropNbl;
 353     }
 354
 355     PCHAR currentOffset = (PCHAR)greHdr + sizeof *greHdr;
 356
 357     if (greHdr->flags & GRE_CSUM) {
 358         tunKey->flags |= OVS_TNL_F_CSUM;
 359         currentOffset += 4;
 360         headRoom += 4;
 361     }
 362
 363     if (greHdr->flags & GRE_KEY) {
 364         tunKey->flags |= OVS_TNL_F_KEY;
 365         UINT32 key = 0;
 366         RtlCopyMemory(&key, currentOffset, 4);
 367         tunKey->tunnelId = (UINT64)key << 32;
 368         currentOffset += 4;
 369         headRoom += 4;
 370     }
 371
 372     /* Clear out the receive flag for the inner packet. */
 373     NET_BUFFER_LIST_INFO(curNbl, TcpIpChecksumNetBufferListInfo) = 0;
 374     NdisAdvanceNetBufferDataStart(curNb, GreTunHdrSize(tunKey->flags), FALSE,
 375                                   NULL);
 376     ASSERT(headRoom == GreTunHdrSize(tunKey->flags));
 377     return NDIS_STATUS_SUCCESS;
 378
 379 dropNbl:
 380     OvsCompleteNBL(switchContext, *newNbl, TRUE);
 381     *newNbl = NULL;
 382     return status;
 383 }