summaryrefslogtreecommitdiffstats
path: root/NetworkPkg/Ip4Dxe/Ip4Input.c
diff options
context:
space:
mode:
Diffstat (limited to 'NetworkPkg/Ip4Dxe/Ip4Input.c')
-rw-r--r--NetworkPkg/Ip4Dxe/Ip4Input.c1597
1 files changed, 1597 insertions, 0 deletions
diff --git a/NetworkPkg/Ip4Dxe/Ip4Input.c b/NetworkPkg/Ip4Dxe/Ip4Input.c
new file mode 100644
index 0000000000..24c5846588
--- /dev/null
+++ b/NetworkPkg/Ip4Dxe/Ip4Input.c
@@ -0,0 +1,1597 @@
+/** @file
+ IP4 input process.
+
+Copyright (c) 2005 - 2018, Intel Corporation. All rights reserved.<BR>
+(C) Copyright 2015 Hewlett-Packard Development Company, L.P.<BR>
+
+SPDX-License-Identifier: BSD-2-Clause-Patent
+
+**/
+
+#include "Ip4Impl.h"
+
+
+/**
+ Create an empty assemble entry for the packet identified by
+ (Dst, Src, Id, Protocol). The default life for the packet is
+ 120 seconds.
+
+ @param[in] Dst The destination address
+ @param[in] Src The source address
+ @param[in] Id The ID field in IP header
+ @param[in] Protocol The protocol field in IP header
+
+ @return NULL if failed to allocate memory for the entry, otherwise
+ the point to just created reassemble entry.
+
+**/
+IP4_ASSEMBLE_ENTRY *
+Ip4CreateAssembleEntry (
+ IN IP4_ADDR Dst,
+ IN IP4_ADDR Src,
+ IN UINT16 Id,
+ IN UINT8 Protocol
+ )
+{
+
+ IP4_ASSEMBLE_ENTRY *Assemble;
+
+ Assemble = AllocatePool (sizeof (IP4_ASSEMBLE_ENTRY));
+
+ if (Assemble == NULL) {
+ return NULL;
+ }
+
+ InitializeListHead (&Assemble->Link);
+ InitializeListHead (&Assemble->Fragments);
+
+ Assemble->Dst = Dst;
+ Assemble->Src = Src;
+ Assemble->Id = Id;
+ Assemble->Protocol = Protocol;
+ Assemble->TotalLen = 0;
+ Assemble->CurLen = 0;
+ Assemble->Head = NULL;
+ Assemble->Info = NULL;
+ Assemble->Life = IP4_FRAGMENT_LIFE;
+
+ return Assemble;
+}
+
+
+/**
+ Release all the fragments of a packet, then free the assemble entry.
+
+ @param[in] Assemble The assemble entry to free
+
+**/
+VOID
+Ip4FreeAssembleEntry (
+ IN IP4_ASSEMBLE_ENTRY *Assemble
+ )
+{
+ LIST_ENTRY *Entry;
+ LIST_ENTRY *Next;
+ NET_BUF *Fragment;
+
+ NET_LIST_FOR_EACH_SAFE (Entry, Next, &Assemble->Fragments) {
+ Fragment = NET_LIST_USER_STRUCT (Entry, NET_BUF, List);
+
+ RemoveEntryList (Entry);
+ NetbufFree (Fragment);
+ }
+
+ FreePool (Assemble);
+}
+
+
+/**
+ Initialize an already allocated assemble table. This is generally
+ the assemble table embedded in the IP4 service instance.
+
+ @param[in, out] Table The assemble table to initialize.
+
+**/
+VOID
+Ip4InitAssembleTable (
+ IN OUT IP4_ASSEMBLE_TABLE *Table
+ )
+{
+ UINT32 Index;
+
+ for (Index = 0; Index < IP4_ASSEMLE_HASH_SIZE; Index++) {
+ InitializeListHead (&Table->Bucket[Index]);
+ }
+}
+
+
+/**
+ Clean up the assemble table: remove all the fragments
+ and assemble entries.
+
+ @param[in] Table The assemble table to clean up
+
+**/
+VOID
+Ip4CleanAssembleTable (
+ IN IP4_ASSEMBLE_TABLE *Table
+ )
+{
+ LIST_ENTRY *Entry;
+ LIST_ENTRY *Next;
+ IP4_ASSEMBLE_ENTRY *Assemble;
+ UINT32 Index;
+
+ for (Index = 0; Index < IP4_ASSEMLE_HASH_SIZE; Index++) {
+ NET_LIST_FOR_EACH_SAFE (Entry, Next, &Table->Bucket[Index]) {
+ Assemble = NET_LIST_USER_STRUCT (Entry, IP4_ASSEMBLE_ENTRY, Link);
+
+ RemoveEntryList (Entry);
+ Ip4FreeAssembleEntry (Assemble);
+ }
+ }
+}
+
+
+/**
+ Trim the packet to fit in [Start, End), and update the per
+ packet information.
+
+ @param Packet Packet to trim
+ @param Start The sequence of the first byte to fit in
+ @param End One beyond the sequence of last byte to fit in.
+
+**/
+VOID
+Ip4TrimPacket (
+ IN OUT NET_BUF *Packet,
+ IN INTN Start,
+ IN INTN End
+ )
+{
+ IP4_CLIP_INFO *Info;
+ INTN Len;
+
+ Info = IP4_GET_CLIP_INFO (Packet);
+
+ ASSERT (Info->Start + Info->Length == Info->End);
+ ASSERT ((Info->Start < End) && (Start < Info->End));
+
+ if (Info->Start < Start) {
+ Len = Start - Info->Start;
+
+ NetbufTrim (Packet, (UINT32) Len, NET_BUF_HEAD);
+ Info->Start = Start;
+ Info->Length -= Len;
+ }
+
+ if (End < Info->End) {
+ Len = End - Info->End;
+
+ NetbufTrim (Packet, (UINT32) Len, NET_BUF_TAIL);
+ Info->End = End;
+ Info->Length -= Len;
+ }
+}
+
+
+/**
+ Release all the fragments of the packet. This is the callback for
+ the assembled packet's OnFree. It will free the assemble entry,
+ which in turn will free all the fragments of the packet.
+
+ @param[in] Arg The assemble entry to free
+
+**/
+VOID
+EFIAPI
+Ip4OnFreeFragments (
+ IN VOID *Arg
+ )
+{
+ Ip4FreeAssembleEntry ((IP4_ASSEMBLE_ENTRY *) Arg);
+}
+
+
+/**
+ Reassemble the IP fragments. If all the fragments of the packet
+ have been received, it will wrap the packet in a net buffer then
+ return it to caller. If the packet can't be assembled, NULL is
+ return.
+
+ @param Table The assemble table used. New assemble entry will be created
+ if the Packet is from a new chain of fragments.
+ @param Packet The fragment to assemble. It might be freed if the fragment
+ can't be re-assembled.
+
+ @return NULL if the packet can't be reassemble. The point to just assembled
+ packet if all the fragments of the packet have arrived.
+
+**/
+NET_BUF *
+Ip4Reassemble (
+ IN OUT IP4_ASSEMBLE_TABLE *Table,
+ IN OUT NET_BUF *Packet
+ )
+{
+ IP4_HEAD *IpHead;
+ IP4_CLIP_INFO *This;
+ IP4_CLIP_INFO *Node;
+ IP4_ASSEMBLE_ENTRY *Assemble;
+ LIST_ENTRY *Head;
+ LIST_ENTRY *Prev;
+ LIST_ENTRY *Cur;
+ NET_BUF *Fragment;
+ NET_BUF *NewPacket;
+ INTN Index;
+
+ IpHead = Packet->Ip.Ip4;
+ This = IP4_GET_CLIP_INFO (Packet);
+
+ ASSERT (IpHead != NULL);
+
+ //
+ // First: find the related assemble entry
+ //
+ Assemble = NULL;
+ Index = IP4_ASSEMBLE_HASH (IpHead->Dst, IpHead->Src, IpHead->Id, IpHead->Protocol);
+
+ NET_LIST_FOR_EACH (Cur, &Table->Bucket[Index]) {
+ Assemble = NET_LIST_USER_STRUCT (Cur, IP4_ASSEMBLE_ENTRY, Link);
+
+ if ((Assemble->Dst == IpHead->Dst) && (Assemble->Src == IpHead->Src) &&
+ (Assemble->Id == IpHead->Id) && (Assemble->Protocol == IpHead->Protocol)) {
+ break;
+ }
+ }
+
+ //
+ // Create a new assemble entry if no assemble entry is related to this packet
+ //
+ if (Cur == &Table->Bucket[Index]) {
+ Assemble = Ip4CreateAssembleEntry (
+ IpHead->Dst,
+ IpHead->Src,
+ IpHead->Id,
+ IpHead->Protocol
+ );
+
+ if (Assemble == NULL) {
+ goto DROP;
+ }
+
+ InsertHeadList (&Table->Bucket[Index], &Assemble->Link);
+ }
+ //
+ // Assemble shouldn't be NULL here
+ //
+ ASSERT (Assemble != NULL);
+
+ //
+ // Find the point to insert the packet: before the first
+ // fragment with THIS.Start < CUR.Start. the previous one
+ // has PREV.Start <= THIS.Start < CUR.Start.
+ //
+ Head = &Assemble->Fragments;
+
+ NET_LIST_FOR_EACH (Cur, Head) {
+ Fragment = NET_LIST_USER_STRUCT (Cur, NET_BUF, List);
+
+ if (This->Start < IP4_GET_CLIP_INFO (Fragment)->Start) {
+ break;
+ }
+ }
+
+ //
+ // Check whether the current fragment overlaps with the previous one.
+ // It holds that: PREV.Start <= THIS.Start < THIS.End. Only need to
+ // check whether THIS.Start < PREV.End for overlap. If two fragments
+ // overlaps, trim the overlapped part off THIS fragment.
+ //
+ if ((Prev = Cur->BackLink) != Head) {
+ Fragment = NET_LIST_USER_STRUCT (Prev, NET_BUF, List);
+ Node = IP4_GET_CLIP_INFO (Fragment);
+
+ if (This->Start < Node->End) {
+ if (This->End <= Node->End) {
+ NetbufFree (Packet);
+ return NULL;
+ }
+
+ Ip4TrimPacket (Packet, Node->End, This->End);
+ }
+ }
+
+ //
+ // Insert the fragment into the packet. The fragment may be removed
+ // from the list by the following checks.
+ //
+ NetListInsertBefore (Cur, &Packet->List);
+
+ //
+ // Check the packets after the insert point. It holds that:
+ // THIS.Start <= NODE.Start < NODE.End. The equality holds
+ // if PREV and NEXT are continuous. THIS fragment may fill
+ // several holes. Remove the completely overlapped fragments
+ //
+ while (Cur != Head) {
+ Fragment = NET_LIST_USER_STRUCT (Cur, NET_BUF, List);
+ Node = IP4_GET_CLIP_INFO (Fragment);
+
+ //
+ // Remove fragments completely overlapped by this fragment
+ //
+ if (Node->End <= This->End) {
+ Cur = Cur->ForwardLink;
+
+ RemoveEntryList (&Fragment->List);
+ Assemble->CurLen -= Node->Length;
+
+ NetbufFree (Fragment);
+ continue;
+ }
+
+ //
+ // The conditions are: THIS.Start <= NODE.Start, and THIS.End <
+ // NODE.End. Two fragments overlaps if NODE.Start < THIS.End.
+ // If two fragments start at the same offset, remove THIS fragment
+ // because ((THIS.Start == NODE.Start) && (THIS.End < NODE.End)).
+ //
+ if (Node->Start < This->End) {
+ if (This->Start == Node->Start) {
+ RemoveEntryList (&Packet->List);
+ goto DROP;
+ }
+
+ Ip4TrimPacket (Packet, This->Start, Node->Start);
+ }
+
+ break;
+ }
+
+ //
+ // Update the assemble info: increase the current length. If it is
+ // the frist fragment, update the packet's IP head and per packet
+ // info. If it is the last fragment, update the total length.
+ //
+ Assemble->CurLen += This->Length;
+
+ if (This->Start == 0) {
+ //
+ // Once the first fragment is enqueued, it can't be removed
+ // from the fragment list. So, Assemble->Head always point
+ // to valid memory area.
+ //
+ ASSERT (Assemble->Head == NULL);
+
+ Assemble->Head = IpHead;
+ Assemble->Info = IP4_GET_CLIP_INFO (Packet);
+ }
+
+ //
+ // Don't update the length more than once.
+ //
+ if (IP4_LAST_FRAGMENT (IpHead->Fragment) && (Assemble->TotalLen == 0)) {
+ Assemble->TotalLen = This->End;
+ }
+
+ //
+ // Deliver the whole packet if all the fragments received.
+ // All fragments received if:
+ // 1. received the last one, so, the total length is know
+ // 2. received all the data. If the last fragment on the
+ // queue ends at the total length, all data is received.
+ //
+ if ((Assemble->TotalLen != 0) && (Assemble->CurLen >= Assemble->TotalLen)) {
+
+ RemoveEntryList (&Assemble->Link);
+
+ //
+ // If the packet is properly formated, the last fragment's End
+ // equals to the packet's total length. Otherwise, the packet
+ // is a fake, drop it now.
+ //
+ Fragment = NET_LIST_USER_STRUCT (Head->BackLink, NET_BUF, List);
+
+ if (IP4_GET_CLIP_INFO (Fragment)->End != Assemble->TotalLen) {
+ Ip4FreeAssembleEntry (Assemble);
+ return NULL;
+ }
+
+ //
+ // Wrap the packet in a net buffer then deliver it up
+ //
+ NewPacket = NetbufFromBufList (
+ &Assemble->Fragments,
+ 0,
+ 0,
+ Ip4OnFreeFragments,
+ Assemble
+ );
+
+ if (NewPacket == NULL) {
+ Ip4FreeAssembleEntry (Assemble);
+ return NULL;
+ }
+
+ NewPacket->Ip.Ip4 = Assemble->Head;
+
+ ASSERT (Assemble->Info != NULL);
+
+ CopyMem (
+ IP4_GET_CLIP_INFO (NewPacket),
+ Assemble->Info,
+ sizeof (*IP4_GET_CLIP_INFO (NewPacket))
+ );
+
+ return NewPacket;
+ }
+
+ return NULL;
+
+DROP:
+ NetbufFree (Packet);
+ return NULL;
+}
+
+/**
+ The callback function for the net buffer which wraps the packet processed by
+ IPsec. It releases the wrap packet and also signals IPsec to free the resources.
+
+ @param[in] Arg The wrap context
+
+**/
+VOID
+EFIAPI
+Ip4IpSecFree (
+ IN VOID *Arg
+ )
+{
+ IP4_IPSEC_WRAP *Wrap;
+
+ Wrap = (IP4_IPSEC_WRAP *) Arg;
+
+ if (Wrap->IpSecRecycleSignal != NULL) {
+ gBS->SignalEvent (Wrap->IpSecRecycleSignal);
+ }
+
+ NetbufFree (Wrap->Packet);
+
+ FreePool (Wrap);
+
+ return;
+}
+
+/**
+ The work function to locate IPsec protocol to process the inbound or
+ outbound IP packets. The process routine handls the packet with following
+ actions: bypass the packet, discard the packet, or protect the packet.
+
+ @param[in] IpSb The IP4 service instance.
+ @param[in, out] Head The The caller supplied IP4 header.
+ @param[in, out] Netbuf The IP4 packet to be processed by IPsec.
+ @param[in, out] Options The caller supplied options.
+ @param[in, out] OptionsLen The length of the option.
+ @param[in] Direction The directionality in an SPD entry,
+ EfiIPsecInBound or EfiIPsecOutBound.
+ @param[in] Context The token's wrap.
+
+ @retval EFI_SUCCESS The IPsec protocol is not available or disabled.
+ @retval EFI_SUCCESS The packet was bypassed and all buffers remain the same.
+ @retval EFI_SUCCESS The packet was protected.
+ @retval EFI_ACCESS_DENIED The packet was discarded.
+ @retval EFI_OUT_OF_RESOURCES There is no suffcient resource to complete the operation.
+ @retval EFI_BUFFER_TOO_SMALL The number of non-empty block is bigger than the
+ number of input data blocks when build a fragment table.
+
+**/
+EFI_STATUS
+Ip4IpSecProcessPacket (
+ IN IP4_SERVICE *IpSb,
+ IN OUT IP4_HEAD **Head,
+ IN OUT NET_BUF **Netbuf,
+ IN OUT UINT8 **Options,
+ IN OUT UINT32 *OptionsLen,
+ IN EFI_IPSEC_TRAFFIC_DIR Direction,
+ IN VOID *Context
+ )
+{
+ NET_FRAGMENT *FragmentTable;
+ NET_FRAGMENT *OriginalFragmentTable;
+ UINT32 FragmentCount;
+ UINT32 OriginalFragmentCount;
+ EFI_EVENT RecycleEvent;
+ NET_BUF *Packet;
+ IP4_TXTOKEN_WRAP *TxWrap;
+ IP4_IPSEC_WRAP *IpSecWrap;
+ EFI_STATUS Status;
+ IP4_HEAD ZeroHead;
+
+ Status = EFI_SUCCESS;
+
+ if (!mIpSec2Installed) {
+ goto ON_EXIT;
+ }
+ ASSERT (mIpSec != NULL);
+
+ Packet = *Netbuf;
+ RecycleEvent = NULL;
+ IpSecWrap = NULL;
+ FragmentTable = NULL;
+ TxWrap = (IP4_TXTOKEN_WRAP *) Context;
+ FragmentCount = Packet->BlockOpNum;
+
+ ZeroMem (&ZeroHead, sizeof (IP4_HEAD));
+
+ //
+ // Check whether the IPsec enable variable is set.
+ //
+ if (mIpSec->DisabledFlag) {
+ //
+ // If IPsec is disabled, restore the original MTU
+ //
+ IpSb->MaxPacketSize = IpSb->OldMaxPacketSize;
+ goto ON_EXIT;
+ } else {
+ //
+ // If IPsec is enabled, use the MTU which reduce the IPsec header length.
+ //
+ IpSb->MaxPacketSize = IpSb->OldMaxPacketSize - IP4_MAX_IPSEC_HEADLEN;
+ }
+
+ //
+ // Rebuild fragment table from netbuf to ease IPsec process.
+ //
+ FragmentTable = AllocateZeroPool (FragmentCount * sizeof (NET_FRAGMENT));
+
+ if (FragmentTable == NULL) {
+ Status = EFI_OUT_OF_RESOURCES;
+ goto ON_EXIT;
+ }
+
+ Status = NetbufBuildExt (Packet, FragmentTable, &FragmentCount);
+
+ //
+ // Record the original FragmentTable and count.
+ //
+ OriginalFragmentTable = FragmentTable;
+ OriginalFragmentCount = FragmentCount;
+
+ if (EFI_ERROR (Status)) {
+ FreePool (FragmentTable);
+ goto ON_EXIT;
+ }
+
+ //
+ // Convert host byte order to network byte order
+ //
+ Ip4NtohHead (*Head);
+
+ Status = mIpSec->ProcessExt (
+ mIpSec,
+ IpSb->Controller,
+ IP_VERSION_4,
+ (VOID *) (*Head),
+ &(*Head)->Protocol,
+ (VOID **) Options,
+ OptionsLen,
+ (EFI_IPSEC_FRAGMENT_DATA **) (&FragmentTable),
+ &FragmentCount,
+ Direction,
+ &RecycleEvent
+ );
+ //
+ // Convert back to host byte order
+ //
+ Ip4NtohHead (*Head);
+
+ if (EFI_ERROR (Status)) {
+ FreePool (OriginalFragmentTable);
+ goto ON_EXIT;
+ }
+
+ if (OriginalFragmentTable == FragmentTable && OriginalFragmentCount == FragmentCount) {
+ //
+ // For ByPass Packet
+ //
+ FreePool (FragmentTable);
+ goto ON_EXIT;
+ } else {
+ //
+ // Free the FragmentTable which allocated before calling the IPsec.
+ //
+ FreePool (OriginalFragmentTable);
+ }
+
+ if (Direction == EfiIPsecOutBound && TxWrap != NULL) {
+
+ TxWrap->IpSecRecycleSignal = RecycleEvent;
+ TxWrap->Packet = NetbufFromExt (
+ FragmentTable,
+ FragmentCount,
+ IP4_MAX_HEADLEN,
+ 0,
+ Ip4FreeTxToken,
+ TxWrap
+ );
+ if (TxWrap->Packet == NULL) {
+ //
+ // Recover the TxWrap->Packet, if meet a error, and the caller will free
+ // the TxWrap.
+ //
+ TxWrap->Packet = *Netbuf;
+ Status = EFI_OUT_OF_RESOURCES;
+ goto ON_EXIT;
+ }
+
+ //
+ // Free orginal Netbuf.
+ //
+ NetIpSecNetbufFree (*Netbuf);
+ *Netbuf = TxWrap->Packet;
+
+ } else {
+
+ IpSecWrap = AllocateZeroPool (sizeof (IP4_IPSEC_WRAP));
+
+ if (IpSecWrap == NULL) {
+ Status = EFI_OUT_OF_RESOURCES;
+ gBS->SignalEvent (RecycleEvent);
+ goto ON_EXIT;
+ }
+
+ IpSecWrap->IpSecRecycleSignal = RecycleEvent;
+ IpSecWrap->Packet = Packet;
+ Packet = NetbufFromExt (
+ FragmentTable,
+ FragmentCount,
+ IP4_MAX_HEADLEN,
+ 0,
+ Ip4IpSecFree,
+ IpSecWrap
+ );
+
+ if (Packet == NULL) {
+ Packet = IpSecWrap->Packet;
+ gBS->SignalEvent (RecycleEvent);
+ FreePool (IpSecWrap);
+ Status = EFI_OUT_OF_RESOURCES;
+ goto ON_EXIT;
+ }
+
+ if (Direction == EfiIPsecInBound && 0 != CompareMem (*Head, &ZeroHead, sizeof (IP4_HEAD))) {
+ Ip4PrependHead (Packet, *Head, *Options, *OptionsLen);
+ Ip4NtohHead (Packet->Ip.Ip4);
+ NetbufTrim (Packet, ((*Head)->HeadLen << 2), TRUE);
+
+ CopyMem (
+ IP4_GET_CLIP_INFO (Packet),
+ IP4_GET_CLIP_INFO (IpSecWrap->Packet),
+ sizeof (IP4_CLIP_INFO)
+ );
+ }
+ *Netbuf = Packet;
+ }
+
+ON_EXIT:
+ return Status;
+}
+
+/**
+ Pre-process the IPv4 packet. First validates the IPv4 packet, and
+ then reassembles packet if it is necessary.
+
+ @param[in] IpSb Pointer to IP4_SERVICE.
+ @param[in, out] Packet Pointer to the Packet to be processed.
+ @param[in] Head Pointer to the IP4_HEAD.
+ @param[in] Option Pointer to a buffer which contains the IPv4 option.
+ @param[in] OptionLen The length of Option in bytes.
+ @param[in] Flag The link layer flag for the packet received, such
+ as multicast.
+
+ @retval EFI_SEUCCESS The recieved packet is in well form.
+ @retval EFI_INVAILD_PARAMETER The recieved packet is malformed.
+
+**/
+EFI_STATUS
+Ip4PreProcessPacket (
+ IN IP4_SERVICE *IpSb,
+ IN OUT NET_BUF **Packet,
+ IN IP4_HEAD *Head,
+ IN UINT8 *Option,
+ IN UINT32 OptionLen,
+ IN UINT32 Flag
+ )
+{
+ IP4_CLIP_INFO *Info;
+ UINT32 HeadLen;
+ UINT32 TotalLen;
+ UINT16 Checksum;
+
+ //
+ // Check if the IP4 header is correctly formatted.
+ //
+ if ((*Packet)->TotalSize < IP4_MIN_HEADLEN) {
+ return EFI_INVALID_PARAMETER;
+ }
+
+ HeadLen = (Head->HeadLen << 2);
+ TotalLen = NTOHS (Head->TotalLen);
+
+ //
+ // Mnp may deliver frame trailer sequence up, trim it off.
+ //
+ if (TotalLen < (*Packet)->TotalSize) {
+ NetbufTrim (*Packet, (*Packet)->TotalSize - TotalLen, FALSE);
+ }
+
+ if ((Head->Ver != 4) || (HeadLen < IP4_MIN_HEADLEN) ||
+ (TotalLen < HeadLen) || (TotalLen != (*Packet)->TotalSize)) {
+ return EFI_INVALID_PARAMETER;
+ }
+
+ //
+ // Some OS may send IP packets without checksum.
+ //
+ Checksum = (UINT16) (~NetblockChecksum ((UINT8 *) Head, HeadLen));
+
+ if ((Head->Checksum != 0) && (Checksum != 0)) {
+ return EFI_INVALID_PARAMETER;
+ }
+
+ //
+ // Convert the IP header to host byte order, then get the per packet info.
+ //
+ (*Packet)->Ip.Ip4 = Ip4NtohHead (Head);
+
+ Info = IP4_GET_CLIP_INFO (*Packet);
+ Info->LinkFlag = Flag;
+ Info->CastType = Ip4GetHostCast (IpSb, Head->Dst, Head->Src);
+ Info->Start = (Head->Fragment & IP4_HEAD_OFFSET_MASK) << 3;
+ Info->Length = Head->TotalLen - HeadLen;
+ Info->End = Info->Start + Info->Length;
+ Info->Status = EFI_SUCCESS;
+
+ //
+ // The packet is destinated to us if the CastType is non-zero.
+ //
+ if ((Info->CastType == 0) || (Info->End > IP4_MAX_PACKET_SIZE)) {
+ return EFI_INVALID_PARAMETER;
+ }
+
+ //
+ // Validate the options. Don't call the Ip4OptionIsValid if
+ // there is no option to save some CPU process.
+ //
+
+ if ((OptionLen > 0) && !Ip4OptionIsValid (Option, OptionLen, TRUE)) {
+ return EFI_INVALID_PARAMETER;
+ }
+
+ //
+ // Trim the head off, after this point, the packet is headless,
+ // and Packet->TotalLen == Info->Length.
+ //
+ NetbufTrim (*Packet, HeadLen, TRUE);
+
+ //
+ // Reassemble the packet if this is a fragment. The packet is a
+ // fragment if its head has MF (more fragment) set, or it starts
+ // at non-zero byte.
+ //
+ if (((Head->Fragment & IP4_HEAD_MF_MASK) != 0) || (Info->Start != 0)) {
+ //
+ // Drop the fragment if DF is set but it is fragmented. Gateway
+ // need to send a type 4 destination unreache ICMP message here.
+ //
+ if ((Head->Fragment & IP4_HEAD_DF_MASK) != 0) {
+ return EFI_INVALID_PARAMETER;
+ }
+
+ //
+ // The length of all but the last fragments is in the unit of 8 bytes.
+ //
+ if (((Head->Fragment & IP4_HEAD_MF_MASK) != 0) && (Info->Length % 8 != 0)) {
+ return EFI_INVALID_PARAMETER;
+ }
+
+ *Packet = Ip4Reassemble (&IpSb->Assemble, *Packet);
+
+ //
+ // Packet assembly isn't complete, start receive more packet.
+ //
+ if (*Packet == NULL) {
+ return EFI_INVALID_PARAMETER;
+ }
+ }
+
+ return EFI_SUCCESS;
+}
+
+/**
+ The IP4 input routine. It is called by the IP4_INTERFACE when a
+ IP4 fragment is received from MNP.
+
+ @param[in] Ip4Instance The IP4 child that request the receive, most like
+ it is NULL.
+ @param[in] Packet The IP4 packet received.
+ @param[in] IoStatus The return status of receive request.
+ @param[in] Flag The link layer flag for the packet received, such
+ as multicast.
+ @param[in] Context The IP4 service instance that own the MNP.
+
+**/
+VOID
+Ip4AccpetFrame (
+ IN IP4_PROTOCOL *Ip4Instance,
+ IN NET_BUF *Packet,
+ IN EFI_STATUS IoStatus,
+ IN UINT32 Flag,
+ IN VOID *Context
+ )
+{
+ IP4_SERVICE *IpSb;
+ IP4_HEAD *Head;
+ EFI_STATUS Status;
+ IP4_HEAD ZeroHead;
+ UINT8 *Option;
+ UINT32 OptionLen;
+
+ IpSb = (IP4_SERVICE *) Context;
+ Option = NULL;
+
+ if (EFI_ERROR (IoStatus) || (IpSb->State == IP4_SERVICE_DESTROY)) {
+ goto DROP;
+ }
+
+ Head = (IP4_HEAD *) NetbufGetByte (Packet, 0, NULL);
+ ASSERT (Head != NULL);
+ OptionLen = (Head->HeadLen << 2) - IP4_MIN_HEADLEN;
+ if (OptionLen > 0) {
+ Option = (UINT8 *) (Head + 1);
+ }
+
+ //
+ // Validate packet format and reassemble packet if it is necessary.
+ //
+ Status = Ip4PreProcessPacket (
+ IpSb,
+ &Packet,
+ Head,
+ Option,
+ OptionLen,
+ Flag
+ );
+
+ if (EFI_ERROR (Status)) {
+ goto RESTART;
+ }
+
+ //
+ // After trim off, the packet is a esp/ah/udp/tcp/icmp6 net buffer,
+ // and no need consider any other ahead ext headers.
+ //
+ Status = Ip4IpSecProcessPacket (
+ IpSb,
+ &Head,
+ &Packet,
+ &Option,
+ &OptionLen,
+ EfiIPsecInBound,
+ NULL
+ );
+
+ if (EFI_ERROR (Status)) {
+ goto RESTART;
+ }
+
+ //
+ // If the packet is protected by tunnel mode, parse the inner Ip Packet.
+ //
+ ZeroMem (&ZeroHead, sizeof (IP4_HEAD));
+ if (0 == CompareMem (Head, &ZeroHead, sizeof (IP4_HEAD))) {
+ // Packet may have been changed. Head, HeadLen, TotalLen, and
+ // info must be reloaded bofore use. The ownership of the packet
+ // is transfered to the packet process logic.
+ //
+ Head = (IP4_HEAD *) NetbufGetByte (Packet, 0, NULL);
+ ASSERT (Head != NULL);
+ Status = Ip4PreProcessPacket (
+ IpSb,
+ &Packet,
+ Head,
+ Option,
+ OptionLen,
+ Flag
+ );
+ if (EFI_ERROR (Status)) {
+ goto RESTART;
+ }
+ }
+
+ ASSERT (Packet != NULL);
+ Head = Packet->Ip.Ip4;
+ IP4_GET_CLIP_INFO (Packet)->Status = EFI_SUCCESS;
+
+ switch (Head->Protocol) {
+ case EFI_IP_PROTO_ICMP:
+ Ip4IcmpHandle (IpSb, Head, Packet);
+ break;
+
+ case IP4_PROTO_IGMP:
+ Ip4IgmpHandle (IpSb, Head, Packet);
+ break;
+
+ default:
+ Ip4Demultiplex (IpSb, Head, Packet, Option, OptionLen);
+ }
+
+ Packet = NULL;
+
+ //
+ // Dispatch the DPCs queued by the NotifyFunction of the rx token's events
+ // which are signaled with received data.
+ //
+ DispatchDpc ();
+
+RESTART:
+ Ip4ReceiveFrame (IpSb->DefaultInterface, NULL, Ip4AccpetFrame, IpSb);
+
+DROP:
+ if (Packet != NULL) {
+ NetbufFree (Packet);
+ }
+
+ return ;
+}
+
+
+/**
+ Check whether this IP child accepts the packet.
+
+ @param[in] IpInstance The IP child to check
+ @param[in] Head The IP header of the packet
+ @param[in] Packet The data of the packet
+
+ @retval TRUE If the child wants to receive the packet.
+ @retval FALSE Otherwise.
+
+**/
+BOOLEAN
+Ip4InstanceFrameAcceptable (
+ IN IP4_PROTOCOL *IpInstance,
+ IN IP4_HEAD *Head,
+ IN NET_BUF *Packet
+ )
+{
+ IP4_ICMP_ERROR_HEAD Icmp;
+ EFI_IP4_CONFIG_DATA *Config;
+ IP4_CLIP_INFO *Info;
+ UINT16 Proto;
+ UINT32 Index;
+
+ Config = &IpInstance->ConfigData;
+
+ //
+ // Dirty trick for the Tiano UEFI network stack implmentation. If
+ // ReceiveTimeout == -1, the receive of the packet for this instance
+ // is disabled. The UEFI spec don't have such capability. We add
+ // this to improve the performance because IP will make a copy of
+ // the received packet for each accepting instance. Some IP instances
+ // used by UDP/TCP only send packets, they don't wants to receive.
+ //
+ if (Config->ReceiveTimeout == (UINT32)(-1)) {
+ return FALSE;
+ }
+
+ if (Config->AcceptPromiscuous) {
+ return TRUE;
+ }
+
+ //
+ // Use protocol from the IP header embedded in the ICMP error
+ // message to filter, instead of ICMP itself. ICMP handle will
+ // call Ip4Demultiplex to deliver ICMP errors.
+ //
+ Proto = Head->Protocol;
+
+ if ((Proto == EFI_IP_PROTO_ICMP) && (!Config->AcceptAnyProtocol) && (Proto != Config->DefaultProtocol)) {
+ NetbufCopy (Packet, 0, sizeof (Icmp.Head), (UINT8 *) &Icmp.Head);
+
+ if (mIcmpClass[Icmp.Head.Type].IcmpClass == ICMP_ERROR_MESSAGE) {
+ if (!Config->AcceptIcmpErrors) {
+ return FALSE;
+ }
+
+ NetbufCopy (Packet, 0, sizeof (Icmp), (UINT8 *) &Icmp);
+ Proto = Icmp.IpHead.Protocol;
+ }
+ }
+
+ //
+ // Match the protocol
+ //
+ if (!Config->AcceptAnyProtocol && (Proto != Config->DefaultProtocol)) {
+ return FALSE;
+ }
+
+ //
+ // Check for broadcast, the caller has computed the packet's
+ // cast type for this child's interface.
+ //
+ Info = IP4_GET_CLIP_INFO (Packet);
+
+ if (IP4_IS_BROADCAST (Info->CastType)) {
+ return Config->AcceptBroadcast;
+ }
+
+ //
+ // If it is a multicast packet, check whether we are in the group.
+ //
+ if (Info->CastType == IP4_MULTICAST) {
+ //
+ // Receive the multicast if the instance wants to receive all packets.
+ //
+ if (!IpInstance->ConfigData.UseDefaultAddress && (IpInstance->Interface->Ip == 0)) {
+ return TRUE;
+ }
+
+ for (Index = 0; Index < IpInstance->GroupCount; Index++) {
+ if (IpInstance->Groups[Index] == HTONL (Head->Dst)) {
+ break;
+ }
+ }
+
+ return (BOOLEAN)(Index < IpInstance->GroupCount);
+ }
+
+ return TRUE;
+}
+
+
+/**
+ Enqueue a shared copy of the packet to the IP4 child if the
+ packet is acceptable to it. Here the data of the packet is
+ shared, but the net buffer isn't.
+
+ @param[in] IpInstance The IP4 child to enqueue the packet to
+ @param[in] Head The IP header of the received packet
+ @param[in] Packet The data of the received packet
+
+ @retval EFI_NOT_STARTED The IP child hasn't been configured.
+ @retval EFI_INVALID_PARAMETER The child doesn't want to receive the packet
+ @retval EFI_OUT_OF_RESOURCES Failed to allocate some resource
+ @retval EFI_SUCCESS A shared copy the packet is enqueued to the child.
+
+**/
+EFI_STATUS
+Ip4InstanceEnquePacket (
+ IN IP4_PROTOCOL *IpInstance,
+ IN IP4_HEAD *Head,
+ IN NET_BUF *Packet
+ )
+{
+ IP4_CLIP_INFO *Info;
+ NET_BUF *Clone;
+
+ //
+ // Check whether the packet is acceptable to this instance.
+ //
+ if (IpInstance->State != IP4_STATE_CONFIGED) {
+ return EFI_NOT_STARTED;
+ }
+
+ if (!Ip4InstanceFrameAcceptable (IpInstance, Head, Packet)) {
+ return EFI_INVALID_PARAMETER;
+ }
+
+ //
+ // Enque a shared copy of the packet.
+ //
+ Clone = NetbufClone (Packet);
+
+ if (Clone == NULL) {
+ return EFI_OUT_OF_RESOURCES;
+ }
+
+ //
+ // Set the receive time out for the assembled packet. If it expires,
+ // packet will be removed from the queue.
+ //
+ Info = IP4_GET_CLIP_INFO (Clone);
+ Info->Life = IP4_US_TO_SEC (IpInstance->ConfigData.ReceiveTimeout);
+
+ InsertTailList (&IpInstance->Received, &Clone->List);
+ return EFI_SUCCESS;
+}
+
+
+/**
+ The signal handle of IP4's recycle event. It is called back
+ when the upper layer release the packet.
+
+ @param Event The IP4's recycle event.
+ @param Context The context of the handle, which is a
+ IP4_RXDATA_WRAP
+
+**/
+VOID
+EFIAPI
+Ip4OnRecyclePacket (
+ IN EFI_EVENT Event,
+ IN VOID *Context
+ )
+{
+ IP4_RXDATA_WRAP *Wrap;
+
+ Wrap = (IP4_RXDATA_WRAP *) Context;
+
+ EfiAcquireLockOrFail (&Wrap->IpInstance->RecycleLock);
+ RemoveEntryList (&Wrap->Link);
+ EfiReleaseLock (&Wrap->IpInstance->RecycleLock);
+
+ ASSERT (!NET_BUF_SHARED (Wrap->Packet));
+ NetbufFree (Wrap->Packet);
+
+ gBS->CloseEvent (Wrap->RxData.RecycleSignal);
+ FreePool (Wrap);
+}
+
+
+/**
+ Wrap the received packet to a IP4_RXDATA_WRAP, which will be
+ delivered to the upper layer. Each IP4 child that accepts the
+ packet will get a not-shared copy of the packet which is wrapped
+ in the IP4_RXDATA_WRAP. The IP4_RXDATA_WRAP->RxData is passed
+ to the upper layer. Upper layer will signal the recycle event in
+ it when it is done with the packet.
+
+ @param[in] IpInstance The IP4 child to receive the packet.
+ @param[in] Packet The packet to deliver up.
+
+ @retval Wrap if warp the packet succeed.
+ @retval NULL failed to wrap the packet .
+
+**/
+IP4_RXDATA_WRAP *
+Ip4WrapRxData (
+ IN IP4_PROTOCOL *IpInstance,
+ IN NET_BUF *Packet
+ )
+{
+ IP4_RXDATA_WRAP *Wrap;
+ EFI_IP4_RECEIVE_DATA *RxData;
+ EFI_STATUS Status;
+ BOOLEAN RawData;
+
+ Wrap = AllocatePool (IP4_RXDATA_WRAP_SIZE (Packet->BlockOpNum));
+
+ if (Wrap == NULL) {
+ return NULL;
+ }
+
+ InitializeListHead (&Wrap->Link);
+
+ Wrap->IpInstance = IpInstance;
+ Wrap->Packet = Packet;
+ RxData = &Wrap->RxData;
+
+ ZeroMem (RxData, sizeof (EFI_IP4_RECEIVE_DATA));
+
+ Status = gBS->CreateEvent (
+ EVT_NOTIFY_SIGNAL,
+ TPL_NOTIFY,
+ Ip4OnRecyclePacket,
+ Wrap,
+ &RxData->RecycleSignal
+ );
+
+ if (EFI_ERROR (Status)) {
+ FreePool (Wrap);
+ return NULL;
+ }
+
+ ASSERT (Packet->Ip.Ip4 != NULL);
+
+ ASSERT (IpInstance != NULL);
+ RawData = IpInstance->ConfigData.RawData;
+
+ //
+ // The application expects a network byte order header.
+ //
+ if (!RawData) {
+ RxData->HeaderLength = (Packet->Ip.Ip4->HeadLen << 2);
+ RxData->Header = (EFI_IP4_HEADER *) Ip4NtohHead (Packet->Ip.Ip4);
+ RxData->OptionsLength = RxData->HeaderLength - IP4_MIN_HEADLEN;
+ RxData->Options = NULL;
+
+ if (RxData->OptionsLength != 0) {
+ RxData->Options = (VOID *) (RxData->Header + 1);
+ }
+ }
+
+ RxData->DataLength = Packet->TotalSize;
+
+ //
+ // Build the fragment table to be delivered up.
+ //
+ RxData->FragmentCount = Packet->BlockOpNum;
+ NetbufBuildExt (Packet, (NET_FRAGMENT *) RxData->FragmentTable, &RxData->FragmentCount);
+
+ return Wrap;
+}
+
+
+/**
+ Deliver the received packets to upper layer if there are both received
+ requests and enqueued packets. If the enqueued packet is shared, it will
+ duplicate it to a non-shared packet, release the shared packet, then
+ deliver the non-shared packet up.
+
+ @param[in] IpInstance The IP child to deliver the packet up.
+
+ @retval EFI_OUT_OF_RESOURCES Failed to allocate resources to deliver the
+ packets.
+ @retval EFI_SUCCESS All the enqueued packets that can be delivered
+ are delivered up.
+
+**/
+EFI_STATUS
+Ip4InstanceDeliverPacket (
+ IN IP4_PROTOCOL *IpInstance
+ )
+{
+ EFI_IP4_COMPLETION_TOKEN *Token;
+ IP4_RXDATA_WRAP *Wrap;
+ NET_BUF *Packet;
+ NET_BUF *Dup;
+ UINT8 *Head;
+ UINT32 HeadLen;
+
+ //
+ // Deliver a packet if there are both a packet and a receive token.
+ //
+ while (!IsListEmpty (&IpInstance->Received) &&
+ !NetMapIsEmpty (&IpInstance->RxTokens)) {
+
+ Packet = NET_LIST_HEAD (&IpInstance->Received, NET_BUF, List);
+
+ if (!NET_BUF_SHARED (Packet)) {
+ //
+ // If this is the only instance that wants the packet, wrap it up.
+ //
+ Wrap = Ip4WrapRxData (IpInstance, Packet);
+
+ if (Wrap == NULL) {
+ return EFI_OUT_OF_RESOURCES;
+ }
+
+ RemoveEntryList (&Packet->List);
+
+ } else {
+ //
+ // Create a duplicated packet if this packet is shared
+ //
+ if (IpInstance->ConfigData.RawData) {
+ HeadLen = 0;
+ } else {
+ HeadLen = IP4_MAX_HEADLEN;
+ }
+
+ Dup = NetbufDuplicate (Packet, NULL, HeadLen);
+
+ if (Dup == NULL) {
+ return EFI_OUT_OF_RESOURCES;
+ }
+
+ if (!IpInstance->ConfigData.RawData) {
+ //
+ // Copy the IP head over. The packet to deliver up is
+ // headless. Trim the head off after copy. The IP head
+ // may be not continuous before the data.
+ //
+ Head = NetbufAllocSpace (Dup, IP4_MAX_HEADLEN, NET_BUF_HEAD);
+ ASSERT (Head != NULL);
+
+ Dup->Ip.Ip4 = (IP4_HEAD *) Head;
+
+ CopyMem (Head, Packet->Ip.Ip4, Packet->Ip.Ip4->HeadLen << 2);
+ NetbufTrim (Dup, IP4_MAX_HEADLEN, TRUE);
+ }
+
+ Wrap = Ip4WrapRxData (IpInstance, Dup);
+
+ if (Wrap == NULL) {
+ NetbufFree (Dup);
+ return EFI_OUT_OF_RESOURCES;
+ }
+
+ RemoveEntryList (&Packet->List);
+ NetbufFree (Packet);
+
+ Packet = Dup;
+ }
+
+ //
+ // Insert it into the delivered packet, then get a user's
+ // receive token, pass the wrapped packet up.
+ //
+ EfiAcquireLockOrFail (&IpInstance->RecycleLock);
+ InsertHeadList (&IpInstance->Delivered, &Wrap->Link);
+ EfiReleaseLock (&IpInstance->RecycleLock);
+
+ Token = NetMapRemoveHead (&IpInstance->RxTokens, NULL);
+ Token->Status = IP4_GET_CLIP_INFO (Packet)->Status;
+ Token->Packet.RxData = &Wrap->RxData;
+
+ gBS->SignalEvent (Token->Event);
+ }
+
+ return EFI_SUCCESS;
+}
+
+
+/**
+ Enqueue a received packet to all the IP children that share
+ the same interface.
+
+ @param[in] IpSb The IP4 service instance that receive the packet.
+ @param[in] Head The header of the received packet.
+ @param[in] Packet The data of the received packet.
+ @param[in] Option Point to the IP4 packet header options.
+ @param[in] OptionLen Length of the IP4 packet header options.
+ @param[in] IpIf The interface to enqueue the packet to.
+
+ @return The number of the IP4 children that accepts the packet
+
+**/
+INTN
+Ip4InterfaceEnquePacket (
+ IN IP4_SERVICE *IpSb,
+ IN IP4_HEAD *Head,
+ IN NET_BUF *Packet,
+ IN UINT8 *Option,
+ IN UINT32 OptionLen,
+ IN IP4_INTERFACE *IpIf
+ )
+{
+ IP4_PROTOCOL *IpInstance;
+ IP4_CLIP_INFO *Info;
+ LIST_ENTRY *Entry;
+ INTN Enqueued;
+ INTN LocalType;
+ INTN SavedType;
+
+ //
+ // First, check that the packet is acceptable to this interface
+ // and find the local cast type for the interface. A packet sent
+ // to say 192.168.1.1 should NOT be delliever to 10.0.0.1 unless
+ // promiscuous receiving.
+ //
+ LocalType = 0;
+ Info = IP4_GET_CLIP_INFO (Packet);
+
+ if ((Info->CastType == IP4_MULTICAST) || (Info->CastType == IP4_LOCAL_BROADCAST)) {
+ //
+ // If the CastType is multicast, don't need to filter against
+ // the group address here, Ip4InstanceFrameAcceptable will do
+ // that later.
+ //
+ LocalType = Info->CastType;
+
+ } else {
+ //
+ // Check the destination againist local IP. If the station
+ // address is 0.0.0.0, it means receiving all the IP destined
+ // to local non-zero IP. Otherwise, it is necessary to compare
+ // the destination to the interface's IP address.
+ //
+ if (IpIf->Ip == IP4_ALLZERO_ADDRESS) {
+ LocalType = IP4_LOCAL_HOST;
+
+ } else {
+ LocalType = Ip4GetNetCast (Head->Dst, IpIf);
+
+ if ((LocalType == 0) && IpIf->PromiscRecv) {
+ LocalType = IP4_PROMISCUOUS;
+ }
+ }
+ }
+
+ if (LocalType == 0) {
+ return 0;
+ }
+
+ //
+ // Iterate through the ip instances on the interface, enqueue
+ // the packet if filter passed. Save the original cast type,
+ // and pass the local cast type to the IP children on the
+ // interface. The global cast type will be restored later.
+ //
+ SavedType = Info->CastType;
+ Info->CastType = LocalType;
+
+ Enqueued = 0;
+
+ NET_LIST_FOR_EACH (Entry, &IpIf->IpInstances) {
+ IpInstance = NET_LIST_USER_STRUCT (Entry, IP4_PROTOCOL, AddrLink);
+ NET_CHECK_SIGNATURE (IpInstance, IP4_PROTOCOL_SIGNATURE);
+
+ //
+ // In RawData mode, add IPv4 headers and options back to packet.
+ //
+ if ((IpInstance->ConfigData.RawData) && (Option != NULL) && (OptionLen != 0)){
+ Ip4PrependHead (Packet, Head, Option, OptionLen);
+ }
+
+ if (Ip4InstanceEnquePacket (IpInstance, Head, Packet) == EFI_SUCCESS) {
+ Enqueued++;
+ }
+ }
+
+ Info->CastType = SavedType;
+ return Enqueued;
+}
+
+
+/**
+ Deliver the packet for each IP4 child on the interface.
+
+ @param[in] IpSb The IP4 service instance that received the packet
+ @param[in] IpIf The IP4 interface to deliver the packet.
+
+ @retval EFI_SUCCESS It always returns EFI_SUCCESS now
+
+**/
+EFI_STATUS
+Ip4InterfaceDeliverPacket (
+ IN IP4_SERVICE *IpSb,
+ IN IP4_INTERFACE *IpIf
+ )
+{
+ IP4_PROTOCOL *Ip4Instance;
+ LIST_ENTRY *Entry;
+
+ NET_LIST_FOR_EACH (Entry, &IpIf->IpInstances) {
+ Ip4Instance = NET_LIST_USER_STRUCT (Entry, IP4_PROTOCOL, AddrLink);
+ Ip4InstanceDeliverPacket (Ip4Instance);
+ }
+
+ return EFI_SUCCESS;
+}
+
+
+/**
+ Demultiple the packet. the packet delivery is processed in two
+ passes. The first pass will enque a shared copy of the packet
+ to each IP4 child that accepts the packet. The second pass will
+ deliver a non-shared copy of the packet to each IP4 child that
+ has pending receive requests. Data is copied if more than one
+ child wants to consume the packet because each IP child needs
+ its own copy of the packet to make changes.
+
+ @param[in] IpSb The IP4 service instance that received the packet.
+ @param[in] Head The header of the received packet.
+ @param[in] Packet The data of the received packet.
+ @param[in] Option Point to the IP4 packet header options.
+ @param[in] OptionLen Length of the IP4 packet header options.
+
+ @retval EFI_NOT_FOUND No IP child accepts the packet.
+ @retval EFI_SUCCESS The packet is enqueued or delivered to some IP
+ children.
+
+**/
+EFI_STATUS
+Ip4Demultiplex (
+ IN IP4_SERVICE *IpSb,
+ IN IP4_HEAD *Head,
+ IN NET_BUF *Packet,
+ IN UINT8 *Option,
+ IN UINT32 OptionLen
+ )
+{
+ LIST_ENTRY *Entry;
+ IP4_INTERFACE *IpIf;
+ INTN Enqueued;
+
+ //
+ // Two pass delivery: first, enque a shared copy of the packet
+ // to each instance that accept the packet.
+ //
+ Enqueued = 0;
+
+ NET_LIST_FOR_EACH (Entry, &IpSb->Interfaces) {
+ IpIf = NET_LIST_USER_STRUCT (Entry, IP4_INTERFACE, Link);
+
+ if (IpIf->Configured) {
+ Enqueued += Ip4InterfaceEnquePacket (
+ IpSb,
+ Head,
+ Packet,
+ Option,
+ OptionLen,
+ IpIf
+ );
+ }
+ }
+
+ //
+ // Second: deliver a duplicate of the packet to each instance.
+ // Release the local reference first, so that the last instance
+ // getting the packet will not copy the data.
+ //
+ NetbufFree (Packet);
+
+ if (Enqueued == 0) {
+ return EFI_NOT_FOUND;
+ }
+
+ NET_LIST_FOR_EACH (Entry, &IpSb->Interfaces) {
+ IpIf = NET_LIST_USER_STRUCT (Entry, IP4_INTERFACE, Link);
+
+ if (IpIf->Configured) {
+ Ip4InterfaceDeliverPacket (IpSb, IpIf);
+ }
+ }
+
+ return EFI_SUCCESS;
+}
+
+
+/**
+ Timeout the fragment and enqueued packets.
+
+ @param[in] IpSb The IP4 service instance to timeout
+
+**/
+VOID
+Ip4PacketTimerTicking (
+ IN IP4_SERVICE *IpSb
+ )
+{
+ LIST_ENTRY *InstanceEntry;
+ LIST_ENTRY *Entry;
+ LIST_ENTRY *Next;
+ IP4_PROTOCOL *IpInstance;
+ IP4_ASSEMBLE_ENTRY *Assemble;
+ NET_BUF *Packet;
+ IP4_CLIP_INFO *Info;
+ UINT32 Index;
+
+ //
+ // First, time out the fragments. The packet's life is counting down
+ // once the first-arrived fragment was received.
+ //
+ for (Index = 0; Index < IP4_ASSEMLE_HASH_SIZE; Index++) {
+ NET_LIST_FOR_EACH_SAFE (Entry, Next, &IpSb->Assemble.Bucket[Index]) {
+ Assemble = NET_LIST_USER_STRUCT (Entry, IP4_ASSEMBLE_ENTRY, Link);
+
+ if ((Assemble->Life > 0) && (--Assemble->Life == 0)) {
+ RemoveEntryList (Entry);
+ Ip4FreeAssembleEntry (Assemble);
+ }
+ }
+ }
+
+ NET_LIST_FOR_EACH (InstanceEntry, &IpSb->Children) {
+ IpInstance = NET_LIST_USER_STRUCT (InstanceEntry, IP4_PROTOCOL, Link);
+
+ //
+ // Second, time out the assembled packets enqueued on each IP child.
+ //
+ NET_LIST_FOR_EACH_SAFE (Entry, Next, &IpInstance->Received) {
+ Packet = NET_LIST_USER_STRUCT (Entry, NET_BUF, List);
+ Info = IP4_GET_CLIP_INFO (Packet);
+
+ if ((Info->Life > 0) && (--Info->Life == 0)) {
+ RemoveEntryList (Entry);
+ NetbufFree (Packet);
+ }
+ }
+
+ //
+ // Third: time out the transmitted packets.
+ //
+ NetMapIterate (&IpInstance->TxTokens, Ip4SentPacketTicking, NULL);
+ }
+}