audk/NetworkPkg/TcpDxe/TcpInput.c

1708 lines
38 KiB
C

/** @file
TCP input process routines.
Copyright (c) 2009 - 2018, Intel Corporation. All rights reserved.<BR>
SPDX-License-Identifier: BSD-2-Clause-Patent
**/
#include "TcpMain.h"
/**
Check whether the sequence number of the incoming segment is acceptable.
@param[in] Tcb Pointer to the TCP_CB of this TCP instance.
@param[in] Seg Pointer to the incoming segment.
@retval 1 The sequence number is acceptable.
@retval 0 The sequence number is not acceptable.
**/
INTN
TcpSeqAcceptable (
IN TCP_CB *Tcb,
IN TCP_SEG *Seg
)
{
return (TCP_SEQ_LEQ (Tcb->RcvNxt, Seg->End) &&
TCP_SEQ_LT (Seg->Seq, Tcb->RcvWl2 + Tcb->RcvWnd));
}
/**
NewReno fast recovery defined in RFC3782.
@param[in, out] Tcb Pointer to the TCP_CB of this TCP instance.
@param[in] Seg Segment that triggers the fast recovery.
**/
VOID
TcpFastRecover (
IN OUT TCP_CB *Tcb,
IN TCP_SEG *Seg
)
{
UINT32 FlightSize;
UINT32 Acked;
//
// Step 1: Three duplicate ACKs and not in fast recovery
//
if (Tcb->CongestState != TCP_CONGEST_RECOVER) {
//
// Step 1A: Invoking fast retransmission.
//
FlightSize = TCP_SUB_SEQ (Tcb->SndNxt, Tcb->SndUna);
Tcb->Ssthresh = MAX (FlightSize >> 1, (UINT32) (2 * Tcb->SndMss));
Tcb->Recover = Tcb->SndNxt;
Tcb->CongestState = TCP_CONGEST_RECOVER;
TCP_CLEAR_FLG (Tcb->CtrlFlag, TCP_CTRL_RTT_ON);
//
// Step 2: Entering fast retransmission
//
TcpRetransmit (Tcb, Tcb->SndUna);
Tcb->CWnd = Tcb->Ssthresh + 3 * Tcb->SndMss;
DEBUG (
(EFI_D_NET,
"TcpFastRecover: enter fast retransmission for TCB %p, recover point is %d\n",
Tcb,
Tcb->Recover)
);
return;
}
//
// During fast recovery, execute Step 3, 4, 5 of RFC3782
//
if (Seg->Ack == Tcb->SndUna) {
//
// Step 3: Fast Recovery,
// If this is a duplicated ACK, increse Cwnd by SMSS.
//
// Step 4 is skipped here only to be executed later
// by TcpToSendData
//
Tcb->CWnd += Tcb->SndMss;
DEBUG (
(EFI_D_NET,
"TcpFastRecover: received another duplicated ACK (%d) for TCB %p\n",
Seg->Ack,
Tcb)
);
} else {
//
// New data is ACKed, check whether it is a
// full ACK or partial ACK
//
if (TCP_SEQ_GEQ (Seg->Ack, Tcb->Recover)) {
//
// Step 5 - Full ACK:
// deflate the congestion window, and exit fast recovery
//
FlightSize = TCP_SUB_SEQ (Tcb->SndNxt, Tcb->SndUna);
Tcb->CWnd = MIN (Tcb->Ssthresh, FlightSize + Tcb->SndMss);
Tcb->CongestState = TCP_CONGEST_OPEN;
DEBUG (
(EFI_D_NET,
"TcpFastRecover: received a full ACK(%d) for TCB %p, exit fast recovery\n",
Seg->Ack,
Tcb)
);
} else {
//
// Step 5 - Partial ACK:
// fast retransmit the first unacknowledge field
// , then deflate the CWnd
//
TcpRetransmit (Tcb, Seg->Ack);
Acked = TCP_SUB_SEQ (Seg->Ack, Tcb->SndUna);
//
// Deflate the CWnd by the amount of new data
// ACKed by SEG.ACK. If more than one SMSS data
// is ACKed, add back SMSS byte to CWnd after
//
if (Acked >= Tcb->SndMss) {
Acked -= Tcb->SndMss;
}
Tcb->CWnd -= Acked;
DEBUG (
(EFI_D_NET,
"TcpFastRecover: received a partial ACK(%d) for TCB %p\n",
Seg->Ack,
Tcb)
);
}
}
}
/**
NewReno fast loss recovery defined in RFC3792.
@param[in, out] Tcb Pointer to the TCP_CB of this TCP instance.
@param[in] Seg Segment that triggers the fast loss recovery.
**/
VOID
TcpFastLossRecover (
IN OUT TCP_CB *Tcb,
IN TCP_SEG *Seg
)
{
if (TCP_SEQ_GT (Seg->Ack, Tcb->SndUna)) {
//
// New data is ACKed, check whether it is a
// full ACK or partial ACK
//
if (TCP_SEQ_GEQ (Seg->Ack, Tcb->LossRecover)) {
//
// Full ACK: exit the loss recovery.
//
Tcb->LossTimes = 0;
Tcb->CongestState = TCP_CONGEST_OPEN;
DEBUG (
(EFI_D_NET,
"TcpFastLossRecover: received a full ACK(%d) for TCB %p\n",
Seg->Ack,
Tcb)
);
} else {
//
// Partial ACK:
// fast retransmit the first unacknowledge field.
//
TcpRetransmit (Tcb, Seg->Ack);
DEBUG (
(EFI_D_NET,
"TcpFastLossRecover: received a partial ACK(%d) for TCB %p\n",
Seg->Ack,
Tcb)
);
}
}
}
/**
Compute the RTT as specified in RFC2988.
@param[in, out] Tcb Pointer to the TCP_CB of this TCP instance.
@param[in] Measure Currently measured RTT in heartbeats.
**/
VOID
TcpComputeRtt (
IN OUT TCP_CB *Tcb,
IN UINT32 Measure
)
{
INT32 Var;
//
// Step 2.3: Compute the RTO for subsequent RTT measurement.
//
if (Tcb->SRtt != 0) {
Var = Tcb->SRtt - (Measure << TCP_RTT_SHIFT);
if (Var < 0) {
Var = -Var;
}
Tcb->RttVar = (3 * Tcb->RttVar + Var) >> 2;
Tcb->SRtt = 7 * (Tcb->SRtt >> 3) + Measure;
} else {
//
// Step 2.2: compute the first RTT measure
//
Tcb->SRtt = Measure << TCP_RTT_SHIFT;
Tcb->RttVar = Measure << (TCP_RTT_SHIFT - 1);
}
Tcb->Rto = (Tcb->SRtt + MAX (8, 4 * Tcb->RttVar)) >> TCP_RTT_SHIFT;
//
// Step 2.4: Limit the RTO to at least 1 second
// Step 2.5: Limit the RTO to a maximum value that
// is at least 60 second
//
if (Tcb->Rto < TCP_RTO_MIN) {
Tcb->Rto = TCP_RTO_MIN;
} else if (Tcb->Rto > TCP_RTO_MAX) {
Tcb->Rto = TCP_RTO_MAX;
}
DEBUG (
(EFI_D_NET,
"TcpComputeRtt: new RTT for TCB %p computed SRTT: %d RTTVAR: %d RTO: %d\n",
Tcb,
Tcb->SRtt,
Tcb->RttVar,
Tcb->Rto)
);
}
/**
Trim the data; SYN and FIN to fit into the window defined by Left and Right.
@param[in] Nbuf The buffer that contains a received TCP segment without an IP header.
@param[in] Left The sequence number of the window's left edge.
@param[in] Right The sequence number of the window's right edge.
@retval 0 The segment is broken.
@retval 1 The segment is in good shape.
**/
INTN
TcpTrimSegment (
IN NET_BUF *Nbuf,
IN TCP_SEQNO Left,
IN TCP_SEQNO Right
)
{
TCP_SEG *Seg;
TCP_SEQNO Urg;
UINT32 Drop;
Seg = TCPSEG_NETBUF (Nbuf);
//
// If the segment is completely out of window,
// truncate every thing, include SYN and FIN.
//
if (TCP_SEQ_LEQ (Seg->End, Left) || TCP_SEQ_LEQ (Right, Seg->Seq)) {
TCP_CLEAR_FLG (Seg->Flag, TCP_FLG_SYN);
TCP_CLEAR_FLG (Seg->Flag, TCP_FLG_FIN);
Seg->Seq = Seg->End;
NetbufTrim (Nbuf, Nbuf->TotalSize, NET_BUF_HEAD);
return 1;
}
//
// Adjust the buffer header
//
if (TCP_SEQ_LT (Seg->Seq, Left)) {
Drop = TCP_SUB_SEQ (Left, Seg->Seq);
Urg = Seg->Seq + Seg->Urg;
Seg->Seq = Left;
if (TCP_FLG_ON (Seg->Flag, TCP_FLG_SYN)) {
TCP_CLEAR_FLG (Seg->Flag, TCP_FLG_SYN);
Drop--;
}
//
// Adjust the urgent point
//
if (TCP_FLG_ON (Seg->Flag, TCP_FLG_URG)) {
if (TCP_SEQ_LT (Urg, Seg->Seq)) {
TCP_CLEAR_FLG (Seg->Flag, TCP_FLG_URG);
} else {
Seg->Urg = (UINT16) TCP_SUB_SEQ (Urg, Seg->Seq);
}
}
if (Drop != 0) {
NetbufTrim (Nbuf, Drop, NET_BUF_HEAD);
}
}
//
// Adjust the buffer tail
//
if (TCP_SEQ_GT (Seg->End, Right)) {
Drop = TCP_SUB_SEQ (Seg->End, Right);
Seg->End = Right;
if (TCP_FLG_ON (Seg->Flag, TCP_FLG_FIN)) {
TCP_CLEAR_FLG (Seg->Flag, TCP_FLG_FIN);
Drop--;
}
if (Drop != 0) {
NetbufTrim (Nbuf, Drop, NET_BUF_TAIL);
}
}
return TcpVerifySegment (Nbuf);
}
/**
Trim off the data outside the tcb's receive window.
@param[in] Tcb Pointer to the TCP_CB of this TCP instance.
@param[in] Nbuf Pointer to the NET_BUF containing the received tcp segment.
@retval 0 The segment is broken.
@retval 1 The segment is in good shape.
**/
INTN
TcpTrimInWnd (
IN TCP_CB *Tcb,
IN NET_BUF *Nbuf
)
{
return TcpTrimSegment (Nbuf, Tcb->RcvNxt, Tcb->RcvWl2 + Tcb->RcvWnd);
}
/**
Process the data and FIN flag, and check whether to deliver
data to the socket layer.
@param[in, out] Tcb Pointer to the TCP_CB of this TCP instance.
@retval 0 No error occurred to deliver data.
@retval -1 An error condition occurred. The proper response is to reset the
connection.
**/
INTN
TcpDeliverData (
IN OUT TCP_CB *Tcb
)
{
LIST_ENTRY *Entry;
NET_BUF *Nbuf;
TCP_SEQNO Seq;
TCP_SEG *Seg;
UINT32 Urgent;
ASSERT ((Tcb != NULL) && (Tcb->Sk != NULL));
//
// make sure there is some data queued,
// and TCP is in a proper state
//
if (IsListEmpty (&Tcb->RcvQue) || !TCP_CONNECTED (Tcb->State)) {
return 0;
}
//
// Deliver data to the socket layer
//
Entry = Tcb->RcvQue.ForwardLink;
Seq = Tcb->RcvNxt;
while (Entry != &Tcb->RcvQue) {
Nbuf = NET_LIST_USER_STRUCT (Entry, NET_BUF, List);
Seg = TCPSEG_NETBUF (Nbuf);
if (TcpVerifySegment (Nbuf) == 0) {
DEBUG (
(EFI_D_ERROR,
"TcpToSendData: discard a broken segment for TCB %p\n",
Tcb)
);
NetbufFree (Nbuf);
return -1;
}
ASSERT (Nbuf->Tcp == NULL);
if (TCP_SEQ_GT (Seg->Seq, Seq)) {
break;
}
Entry = Entry->ForwardLink;
Seq = Seg->End;
Tcb->RcvNxt = Seq;
RemoveEntryList (&Nbuf->List);
//
// RFC793 Eighth step: process FIN in sequence
//
if (TCP_FLG_ON (Seg->Flag, TCP_FLG_FIN)) {
//
// The peer sends to us junky data after FIN,
// reset the connection.
//
if (!IsListEmpty (&Tcb->RcvQue)) {
DEBUG (
(EFI_D_ERROR,
"TcpDeliverData: data received after FIN from peer of TCB %p, reset connection\n",
Tcb)
);
NetbufFree (Nbuf);
return -1;
}
DEBUG (
(EFI_D_NET,
"TcpDeliverData: processing FIN from peer of TCB %p\n",
Tcb)
);
switch (Tcb->State) {
case TCP_SYN_RCVD:
case TCP_ESTABLISHED:
TcpSetState (Tcb, TCP_CLOSE_WAIT);
break;
case TCP_FIN_WAIT_1:
if (!TCP_FLG_ON (Tcb->CtrlFlag, TCP_CTRL_FIN_ACKED)) {
TcpSetState (Tcb, TCP_CLOSING);
break;
}
//
// fall through
//
case TCP_FIN_WAIT_2:
TcpSetState (Tcb, TCP_TIME_WAIT);
TcpClearAllTimer (Tcb);
if (Tcb->TimeWaitTimeout != 0) {
TcpSetTimer (Tcb, TCP_TIMER_2MSL, Tcb->TimeWaitTimeout);
} else {
DEBUG (
(EFI_D_WARN,
"Connection closed immediately because app disables TIME_WAIT timer for %p\n",
Tcb)
);
TcpSendAck (Tcb);
TcpClose (Tcb);
}
break;
case TCP_CLOSE_WAIT:
case TCP_CLOSING:
case TCP_LAST_ACK:
case TCP_TIME_WAIT:
//
// The peer sends to us junk FIN byte. Discard
// the buffer then reset the connection
//
NetbufFree (Nbuf);
return -1;
break;
default:
break;
}
TCP_SET_FLG (Tcb->CtrlFlag, TCP_CTRL_ACK_NOW);
Seg->End--;
}
//
// Don't delay the ack if PUSH flag is on.
//
if (TCP_FLG_ON (Seg->Flag, TCP_FLG_PSH)) {
TCP_SET_FLG (Tcb->CtrlFlag, TCP_CTRL_ACK_NOW);
}
if (Nbuf->TotalSize != 0) {
Urgent = 0;
if (TCP_FLG_ON (Tcb->CtrlFlag, TCP_CTRL_RCVD_URG) &&
TCP_SEQ_LEQ (Seg->Seq, Tcb->RcvUp))
{
if (TCP_SEQ_LEQ (Seg->End, Tcb->RcvUp)) {
Urgent = Nbuf->TotalSize;
} else {
Urgent = TCP_SUB_SEQ (Tcb->RcvUp, Seg->Seq) + 1;
}
}
SockDataRcvd (Tcb->Sk, Nbuf, Urgent);
}
if (TCP_FIN_RCVD (Tcb->State)) {
SockNoMoreData (Tcb->Sk);
}
NetbufFree (Nbuf);
}
return 0;
}
/**
Store the data into the reassemble queue.
@param[in, out] Tcb Pointer to the TCP_CB of this TCP instance.
@param[in] Nbuf Pointer to the buffer containing the data to be queued.
@retval 0 An error condition occurred.
@retval 1 No error occurred to queue data.
**/
INTN
TcpQueueData (
IN OUT TCP_CB *Tcb,
IN NET_BUF *Nbuf
)
{
TCP_SEG *Seg;
LIST_ENTRY *Head;
LIST_ENTRY *Prev;
LIST_ENTRY *Cur;
NET_BUF *Node;
ASSERT ((Tcb != NULL) && (Nbuf != NULL) && (Nbuf->Tcp == NULL));
NET_GET_REF (Nbuf);
Seg = TCPSEG_NETBUF (Nbuf);
Head = &Tcb->RcvQue;
//
// Fast path to process normal case. That is,
// no out-of-order segments are received.
//
if (IsListEmpty (Head)) {
InsertTailList (Head, &Nbuf->List);
return 1;
}
//
// Find the point to insert the buffer
//
for (Prev = Head, Cur = Head->ForwardLink;
Cur != Head;
Prev = Cur, Cur = Cur->ForwardLink) {
Node = NET_LIST_USER_STRUCT (Cur, NET_BUF, List);
if (TCP_SEQ_LT (Seg->Seq, TCPSEG_NETBUF (Node)->Seq)) {
break;
}
}
//
// Check whether the current segment overlaps with the
// previous segment.
//
if (Prev != Head) {
Node = NET_LIST_USER_STRUCT (Prev, NET_BUF, List);
if (TCP_SEQ_LT (Seg->Seq, TCPSEG_NETBUF (Node)->End)) {
if (TCP_SEQ_LEQ (Seg->End, TCPSEG_NETBUF (Node)->End)) {
return 1;
}
if (TcpTrimSegment (Nbuf, TCPSEG_NETBUF (Node)->End, Seg->End) == 0) {
return 0;
}
}
}
InsertHeadList (Prev, &Nbuf->List);
TCP_SET_FLG (Tcb->CtrlFlag, TCP_CTRL_ACK_NOW);
//
// Check the segments after the insert point.
//
while (Cur != Head) {
Node = NET_LIST_USER_STRUCT (Cur, NET_BUF, List);
if (TCP_SEQ_LEQ (TCPSEG_NETBUF (Node)->End, Seg->End)) {
Cur = Cur->ForwardLink;
RemoveEntryList (&Node->List);
NetbufFree (Node);
continue;
}
if (TCP_SEQ_LT (TCPSEG_NETBUF (Node)->Seq, Seg->End)) {
if (TCP_SEQ_LEQ (TCPSEG_NETBUF (Node)->Seq, Seg->Seq)) {
RemoveEntryList (&Nbuf->List);
return 1;
}
if (TcpTrimSegment (Nbuf, Seg->Seq, TCPSEG_NETBUF (Node)->Seq) == 0) {
RemoveEntryList (&Nbuf->List);
return 0;
}
break;
}
Cur = Cur->ForwardLink;
}
return 1;
}
/**
Adjust the send queue or the retransmit queue.
@param[in] Tcb Pointer to the TCP_CB of this TCP instance.
@param[in] Ack The acknowledge sequence number of the received segment.
@retval 0 An error condition occurred.
@retval 1 No error occurred.
**/
INTN
TcpAdjustSndQue (
IN TCP_CB *Tcb,
IN TCP_SEQNO Ack
)
{
LIST_ENTRY *Head;
LIST_ENTRY *Cur;
NET_BUF *Node;
TCP_SEG *Seg;
Head = &Tcb->SndQue;
Cur = Head->ForwardLink;
while (Cur != Head) {
Node = NET_LIST_USER_STRUCT (Cur, NET_BUF, List);
Seg = TCPSEG_NETBUF (Node);
if (TCP_SEQ_GEQ (Seg->Seq, Ack)) {
break;
}
//
// Remove completely ACKed segments
//
if (TCP_SEQ_LEQ (Seg->End, Ack)) {
Cur = Cur->ForwardLink;
RemoveEntryList (&Node->List);
NetbufFree (Node);
continue;
}
return TcpTrimSegment (Node, Ack, Seg->End);
}
return 1;
}
/**
Process the received TCP segments.
@param[in] Nbuf Buffer that contains received a TCP segment without an IP header.
@param[in] Src Source address of the segment, or the peer's IP address.
@param[in] Dst Destination address of the segment, or the local end's IP
address.
@param[in] Version IP_VERSION_4 indicates IP4 stack. IP_VERSION_6 indicates
IP6 stack.
@retval 0 Segment processed successfully. It is either accepted or
discarded. However, no connection is reset by the segment.
@retval -1 A connection is reset by the segment.
**/
INTN
TcpInput (
IN NET_BUF *Nbuf,
IN EFI_IP_ADDRESS *Src,
IN EFI_IP_ADDRESS *Dst,
IN UINT8 Version
)
{
TCP_CB *Tcb;
TCP_CB *Parent;
TCP_OPTION Option;
TCP_HEAD *Head;
INT32 Len;
TCP_SEG *Seg;
TCP_SEQNO Right;
TCP_SEQNO Urg;
UINT16 Checksum;
INT32 Usable;
ASSERT ((Version == IP_VERSION_4) || (Version == IP_VERSION_6));
NET_CHECK_SIGNATURE (Nbuf, NET_BUF_SIGNATURE);
Parent = NULL;
Tcb = NULL;
Head = (TCP_HEAD *) NetbufGetByte (Nbuf, 0, NULL);
ASSERT (Head != NULL);
if (Nbuf->TotalSize < sizeof (TCP_HEAD)) {
DEBUG ((EFI_D_NET, "TcpInput: received a malformed packet\n"));
goto DISCARD;
}
Len = Nbuf->TotalSize - (Head->HeadLen << 2);
if ((Head->HeadLen < 5) || (Len < 0)) {
DEBUG ((EFI_D_NET, "TcpInput: received a malformed packet\n"));
goto DISCARD;
}
if (Version == IP_VERSION_4) {
Checksum = NetPseudoHeadChecksum (Src->Addr[0], Dst->Addr[0], 6, 0);
} else {
Checksum = NetIp6PseudoHeadChecksum (&Src->v6, &Dst->v6, 6, 0);
}
Checksum = TcpChecksum (Nbuf, Checksum);
if (Checksum != 0) {
DEBUG ((EFI_D_ERROR, "TcpInput: received a checksum error packet\n"));
goto DISCARD;
}
if (TCP_FLG_ON (Head->Flag, TCP_FLG_SYN)) {
Len++;
}
if (TCP_FLG_ON (Head->Flag, TCP_FLG_FIN)) {
Len++;
}
Tcb = TcpLocateTcb (
Head->DstPort,
Dst,
Head->SrcPort,
Src,
Version,
(BOOLEAN) TCP_FLG_ON (Head->Flag, TCP_FLG_SYN)
);
if ((Tcb == NULL) || (Tcb->State == TCP_CLOSED)) {
DEBUG ((EFI_D_NET, "TcpInput: send reset because no TCB found\n"));
Tcb = NULL;
goto SEND_RESET;
}
Seg = TcpFormatNetbuf (Tcb, Nbuf);
//
// RFC1122 recommended reaction to illegal option
// (in fact, an illegal option length) is reset.
//
if (TcpParseOption (Nbuf->Tcp, &Option) == -1) {
DEBUG (
(EFI_D_ERROR,
"TcpInput: reset the peer because of malformed option for TCB %p\n",
Tcb)
);
goto SEND_RESET;
}
//
// From now on, the segment is headless
//
NetbufTrim (Nbuf, (Head->HeadLen << 2), NET_BUF_HEAD);
Nbuf->Tcp = NULL;
//
// Process the segment in LISTEN state.
//
if (Tcb->State == TCP_LISTEN) {
//
// First step: Check RST
//
if (TCP_FLG_ON (Seg->Flag, TCP_FLG_RST)) {
DEBUG (
(EFI_D_WARN,
"TcpInput: discard a reset segment for TCB %p in listening\n",
Tcb)
);
goto DISCARD;
}
//
// Second step: Check ACK.
// Any ACK sent to TCP in LISTEN is reseted.
//
if (TCP_FLG_ON (Seg->Flag, TCP_FLG_ACK)) {
DEBUG (
(EFI_D_WARN,
"TcpInput: send reset because of segment with ACK for TCB %p in listening\n",
Tcb)
);
goto SEND_RESET;
}
//
// Third step: Check SYN
//
if (TCP_FLG_ON (Seg->Flag, TCP_FLG_SYN)) {
//
// create a child TCB to handle the data
//
Parent = Tcb;
Tcb = TcpCloneTcb (Parent);
if (Tcb == NULL) {
DEBUG (
(EFI_D_ERROR,
"TcpInput: discard a segment because failed to clone a child for TCB %p\n",
Tcb)
);
goto DISCARD;
}
DEBUG (
(EFI_D_NET,
"TcpInput: create a child for TCB %p in listening\n",
Tcb)
);
//
// init the TCB structure
//
IP6_COPY_ADDRESS (&Tcb->LocalEnd.Ip, Dst);
IP6_COPY_ADDRESS (&Tcb->RemoteEnd.Ip, Src);
Tcb->LocalEnd.Port = Head->DstPort;
Tcb->RemoteEnd.Port = Head->SrcPort;
TcpInitTcbLocal (Tcb);
TcpInitTcbPeer (Tcb, Seg, &Option);
TcpSetState (Tcb, TCP_SYN_RCVD);
TcpSetTimer (Tcb, TCP_TIMER_CONNECT, Tcb->ConnectTimeout);
if (TcpTrimInWnd (Tcb, Nbuf) == 0) {
DEBUG (
(EFI_D_ERROR,
"TcpInput: discard a broken segment for TCB %p\n",
Tcb)
);
goto DISCARD;
}
goto StepSix;
}
goto DISCARD;
} else if (Tcb->State == TCP_SYN_SENT) {
//
// First step: Check ACK bit
//
if (TCP_FLG_ON (Seg->Flag, TCP_FLG_ACK) && (Seg->Ack != Tcb->Iss + 1)) {
DEBUG (
(EFI_D_WARN,
"TcpInput: send reset because of wrong ACK received for TCB %p in SYN_SENT\n",
Tcb)
);
goto SEND_RESET;
}
//
// Second step: Check RST bit
//
if (TCP_FLG_ON (Seg->Flag, TCP_FLG_RST)) {
if (TCP_FLG_ON (Seg->Flag, TCP_FLG_ACK)) {
DEBUG (
(EFI_D_WARN,
"TcpInput: connection reset by peer for TCB %p in SYN_SENT\n",
Tcb)
);
SOCK_ERROR (Tcb->Sk, EFI_CONNECTION_RESET);
goto DROP_CONNECTION;
} else {
DEBUG (
(EFI_D_WARN,
"TcpInput: discard a reset segment because of no ACK for TCB %p in SYN_SENT\n",
Tcb)
);
goto DISCARD;
}
}
//
// Third step: Check security and precedence. Skipped
//
//
// Fourth step: Check SYN. Pay attention to simultaneous open
//
if (TCP_FLG_ON (Seg->Flag, TCP_FLG_SYN)) {
TcpInitTcbPeer (Tcb, Seg, &Option);
if (TCP_FLG_ON (Seg->Flag, TCP_FLG_ACK)) {
Tcb->SndUna = Seg->Ack;
}
TcpClearTimer (Tcb, TCP_TIMER_REXMIT);
if (TCP_SEQ_GT (Tcb->SndUna, Tcb->Iss)) {
TcpSetState (Tcb, TCP_ESTABLISHED);
TcpClearTimer (Tcb, TCP_TIMER_CONNECT);
TcpDeliverData (Tcb);
if ((Tcb->CongestState == TCP_CONGEST_OPEN) &&
TCP_FLG_ON (Tcb->CtrlFlag, TCP_CTRL_RTT_ON))
{
TcpComputeRtt (Tcb, Tcb->RttMeasure);
TCP_CLEAR_FLG (Tcb->CtrlFlag, TCP_CTRL_RTT_ON);
}
if (TcpTrimInWnd (Tcb, Nbuf) == 0) {
DEBUG (
(EFI_D_ERROR,
"TcpInput: discard a broken segment for TCB %p\n",
Tcb)
);
goto DISCARD;
}
TCP_SET_FLG (Tcb->CtrlFlag, TCP_CTRL_ACK_NOW);
DEBUG (
(EFI_D_NET,
"TcpInput: connection established for TCB %p in SYN_SENT\n",
Tcb)
);
goto StepSix;
} else {
//
// Received a SYN segment without ACK, simultaneous open.
//
TcpSetState (Tcb, TCP_SYN_RCVD);
ASSERT (Tcb->SndNxt == Tcb->Iss + 1);
if (TcpAdjustSndQue (Tcb, Tcb->SndNxt) == 0 || TcpTrimInWnd (Tcb, Nbuf) == 0) {
DEBUG (
(EFI_D_ERROR,
"TcpInput: discard a broken segment for TCB %p\n",
Tcb)
);
goto DISCARD;
}
DEBUG (
(EFI_D_WARN,
"TcpInput: simultaneous open for TCB %p in SYN_SENT\n",
Tcb)
);
goto StepSix;
}
}
goto DISCARD;
}
//
// Process segment in SYN_RCVD or TCP_CONNECTED states
//
//
// Clear probe timer since the RecvWindow is opened.
//
if (Tcb->ProbeTimerOn && (Seg->Wnd != 0)) {
TcpClearTimer (Tcb, TCP_TIMER_PROBE);
Tcb->ProbeTimerOn = FALSE;
}
//
// First step: Check whether SEG.SEQ is acceptable
//
if (TcpSeqAcceptable (Tcb, Seg) == 0) {
DEBUG (
(EFI_D_WARN,
"TcpInput: sequence acceptance test failed for segment of TCB %p\n",
Tcb)
);
if (!TCP_FLG_ON (Seg->Flag, TCP_FLG_RST)) {
TcpSendAck (Tcb);
}
goto DISCARD;
}
if ((TCP_SEQ_LT (Seg->Seq, Tcb->RcvWl2)) &&
(Tcb->RcvWl2 == Seg->End) &&
!TCP_FLG_ON (Seg->Flag, TCP_FLG_SYN | TCP_FLG_FIN))
{
TCP_SET_FLG (Tcb->CtrlFlag, TCP_CTRL_ACK_NOW);
}
//
// Second step: Check the RST
//
if (TCP_FLG_ON (Seg->Flag, TCP_FLG_RST)) {
DEBUG ((EFI_D_WARN, "TcpInput: connection reset for TCB %p\n", Tcb));
if (Tcb->State == TCP_SYN_RCVD) {
SOCK_ERROR (Tcb->Sk, EFI_CONNECTION_REFUSED);
//
// This TCB comes from either a LISTEN TCB,
// or active open TCB with simultaneous open.
// Do NOT signal user CONNECTION refused
// if it comes from a LISTEN TCB.
//
} else if ((Tcb->State == TCP_ESTABLISHED) ||
(Tcb->State == TCP_FIN_WAIT_1) ||
(Tcb->State == TCP_FIN_WAIT_2) ||
(Tcb->State == TCP_CLOSE_WAIT))
{
SOCK_ERROR (Tcb->Sk, EFI_CONNECTION_RESET);
} else {
}
goto DROP_CONNECTION;
}
//
// Trim the data and flags.
//
if (TcpTrimInWnd (Tcb, Nbuf) == 0) {
DEBUG (
(EFI_D_ERROR,
"TcpInput: discard a broken segment for TCB %p\n",
Tcb)
);
goto DISCARD;
}
//
// Third step: Check security and precedence, Ignored
//
//
// Fourth step: Check the SYN bit.
//
if (TCP_FLG_ON (Seg->Flag, TCP_FLG_SYN)) {
DEBUG (
(EFI_D_WARN,
"TcpInput: connection reset because received extra SYN for TCB %p\n",
Tcb)
);
SOCK_ERROR (Tcb->Sk, EFI_CONNECTION_RESET);
goto RESET_THEN_DROP;
}
//
// Fifth step: Check the ACK
//
if (!TCP_FLG_ON (Seg->Flag, TCP_FLG_ACK)) {
DEBUG (
(EFI_D_WARN,
"TcpInput: segment discard because of no ACK for connected TCB %p\n",
Tcb)
);
goto DISCARD;
} else {
if (Tcb->IpInfo->IpVersion == IP_VERSION_6 && Tcb->Tick == 0) {
Tcp6RefreshNeighbor (Tcb, Src, TCP6_KEEP_NEIGHBOR_TIME * TICKS_PER_SECOND);
Tcb->Tick = TCP6_REFRESH_NEIGHBOR_TICK;
}
}
if (Tcb->State == TCP_SYN_RCVD) {
if (TCP_SEQ_LT (Tcb->SndUna, Seg->Ack) &&
TCP_SEQ_LEQ (Seg->Ack, Tcb->SndNxt))
{
Tcb->SndWnd = Seg->Wnd;
Tcb->SndWndMax = MAX (Tcb->SndWnd, Tcb->SndWndMax);
Tcb->SndWl1 = Seg->Seq;
Tcb->SndWl2 = Seg->Ack;
TcpSetState (Tcb, TCP_ESTABLISHED);
TcpClearTimer (Tcb, TCP_TIMER_CONNECT);
TcpDeliverData (Tcb);
DEBUG (
(EFI_D_NET,
"TcpInput: connection established for TCB %p in SYN_RCVD\n",
Tcb)
);
//
// Continue the process as ESTABLISHED state
//
} else {
DEBUG (
(EFI_D_WARN,
"TcpInput: send reset because of wrong ACK for TCB %p in SYN_RCVD\n",
Tcb)
);
goto SEND_RESET;
}
}
if (TCP_SEQ_LT (Seg->Ack, Tcb->SndUna)) {
DEBUG (
(EFI_D_WARN,
"TcpInput: ignore the out-of-data ACK for connected TCB %p\n",
Tcb)
);
goto StepSix;
} else if (TCP_SEQ_GT (Seg->Ack, Tcb->SndNxt)) {
DEBUG (
(EFI_D_WARN,
"TcpInput: discard segment for future ACK for connected TCB %p\n",
Tcb)
);
TcpSendAck (Tcb);
goto DISCARD;
}
//
// From now on: SND.UNA <= SEG.ACK <= SND.NXT.
//
if (TCP_FLG_ON (Option.Flag, TCP_OPTION_RCVD_TS)) {
//
// update TsRecent as specified in page 16 RFC1323.
// RcvWl2 equals to the variable "LastAckSent"
// defined there.
//
if (TCP_SEQ_LEQ (Seg->Seq, Tcb->RcvWl2) &&
TCP_SEQ_LT (Tcb->RcvWl2, Seg->End))
{
Tcb->TsRecent = Option.TSVal;
Tcb->TsRecentAge = mTcpTick;
}
TcpComputeRtt (Tcb, TCP_SUB_TIME (mTcpTick, Option.TSEcr));
} else if (TCP_FLG_ON (Tcb->CtrlFlag, TCP_CTRL_RTT_ON)) {
ASSERT (Tcb->CongestState == TCP_CONGEST_OPEN);
TcpComputeRtt (Tcb, Tcb->RttMeasure);
TCP_CLEAR_FLG (Tcb->CtrlFlag, TCP_CTRL_RTT_ON);
}
if (Seg->Ack == Tcb->SndNxt) {
TcpClearTimer (Tcb, TCP_TIMER_REXMIT);
} else {
TcpSetTimer (Tcb, TCP_TIMER_REXMIT, Tcb->Rto);
}
//
// Count duplicate acks.
//
if ((Seg->Ack == Tcb->SndUna) &&
(Tcb->SndUna != Tcb->SndNxt) &&
(Seg->Wnd == Tcb->SndWnd) &&
(0 == Len))
{
Tcb->DupAck++;
} else {
Tcb->DupAck = 0;
}
//
// Congestion avoidance, fast recovery and fast retransmission.
//
if (((Tcb->CongestState == TCP_CONGEST_OPEN) && (Tcb->DupAck < 3)) ||
(Tcb->CongestState == TCP_CONGEST_LOSS))
{
if (TCP_SEQ_GT (Seg->Ack, Tcb->SndUna)) {
if (Tcb->CWnd < Tcb->Ssthresh) {
Tcb->CWnd += Tcb->SndMss;
} else {
Tcb->CWnd += MAX (Tcb->SndMss * Tcb->SndMss / Tcb->CWnd, 1);
}
Tcb->CWnd = MIN (Tcb->CWnd, TCP_MAX_WIN << Tcb->SndWndScale);
}
if (Tcb->CongestState == TCP_CONGEST_LOSS) {
TcpFastLossRecover (Tcb, Seg);
}
} else {
TcpFastRecover (Tcb, Seg);
}
if (TCP_SEQ_GT (Seg->Ack, Tcb->SndUna)) {
if (TcpAdjustSndQue (Tcb, Seg->Ack) == 0) {
DEBUG (
(EFI_D_ERROR,
"TcpInput: discard a broken segment for TCB %p\n",
Tcb)
);
goto DISCARD;
}
Tcb->SndUna = Seg->Ack;
if (TCP_FLG_ON (Tcb->CtrlFlag, TCP_CTRL_SND_URG) &&
TCP_SEQ_LT (Tcb->SndUp, Seg->Ack))
{
TCP_CLEAR_FLG (Tcb->CtrlFlag, TCP_CTRL_SND_URG);
}
}
//
// Update window info
//
if (TCP_SEQ_LT (Tcb->SndWl1, Seg->Seq) ||
((Tcb->SndWl1 == Seg->Seq) && TCP_SEQ_LEQ (Tcb->SndWl2, Seg->Ack)))
{
Right = Seg->Ack + Seg->Wnd;
if (TCP_SEQ_LT (Right, Tcb->SndWl2 + Tcb->SndWnd)) {
if ((Tcb->SndWl1 == Seg->Seq) &&
(Tcb->SndWl2 == Seg->Ack) &&
(Len == 0))
{
goto NO_UPDATE;
}
DEBUG (
(EFI_D_WARN,
"TcpInput: peer shrinks the window for connected TCB %p\n",
Tcb)
);
if ((Tcb->CongestState == TCP_CONGEST_RECOVER) &&
(TCP_SEQ_LT (Right, Tcb->Recover)))
{
Tcb->Recover = Right;
}
if ((Tcb->CongestState == TCP_CONGEST_LOSS) &&
(TCP_SEQ_LT (Right, Tcb->LossRecover)))
{
Tcb->LossRecover = Right;
}
if (TCP_SEQ_LT (Right, Tcb->SndNxt)) {
//
// Check for Window Retraction in RFC7923 section 2.4.
// The lower n bits of the peer's actual receive window is wiped out if TCP
// window scale is enabled, it will look like the peer is shrinking the window.
// Check whether the SndNxt is out of the advertised receive window by more than
// 2^Rcv.Wind.Shift before moving the SndNxt to the left.
//
DEBUG (
(EFI_D_WARN,
"TcpInput: peer advise negative useable window for connected TCB %p\n",
Tcb)
);
Usable = TCP_SUB_SEQ (Tcb->SndNxt, Right);
if ((Usable >> Tcb->SndWndScale) > 0) {
DEBUG (
(EFI_D_WARN,
"TcpInput: SndNxt is out of window by more than window scale for TCB %p\n",
Tcb)
);
Tcb->SndNxt = Right;
}
if (Right == Tcb->SndUna) {
TcpClearTimer (Tcb, TCP_TIMER_REXMIT);
TcpSetProbeTimer (Tcb);
}
}
}
Tcb->SndWnd = Seg->Wnd;
Tcb->SndWndMax = MAX (Tcb->SndWnd, Tcb->SndWndMax);
Tcb->SndWl1 = Seg->Seq;
Tcb->SndWl2 = Seg->Ack;
}
NO_UPDATE:
if (TCP_FLG_ON (Tcb->CtrlFlag, TCP_CTRL_FIN_SENT) &&
(Tcb->SndUna == Tcb->SndNxt))
{
DEBUG (
(EFI_D_NET,
"TcpInput: local FIN is ACKed by peer for connected TCB %p\n",
Tcb)
);
TCP_SET_FLG (Tcb->CtrlFlag, TCP_CTRL_FIN_ACKED);
}
//
// Transit the state if proper.
//
switch (Tcb->State) {
case TCP_FIN_WAIT_1:
if (TCP_FLG_ON (Tcb->CtrlFlag, TCP_CTRL_FIN_ACKED)) {
TcpSetState (Tcb, TCP_FIN_WAIT_2);
TcpClearAllTimer (Tcb);
TcpSetTimer (Tcb, TCP_TIMER_FINWAIT2, Tcb->FinWait2Timeout);
}
case TCP_FIN_WAIT_2:
break;
case TCP_CLOSE_WAIT:
break;
case TCP_CLOSING:
if (TCP_FLG_ON (Tcb->CtrlFlag, TCP_CTRL_FIN_ACKED)) {
TcpSetState (Tcb, TCP_TIME_WAIT);
TcpClearAllTimer (Tcb);
if (Tcb->TimeWaitTimeout != 0) {
TcpSetTimer (Tcb, TCP_TIMER_2MSL, Tcb->TimeWaitTimeout);
} else {
DEBUG (
(EFI_D_WARN,
"Connection closed immediately because app disables TIME_WAIT timer for %p\n",
Tcb)
);
TcpClose (Tcb);
}
}
break;
case TCP_LAST_ACK:
if (TCP_FLG_ON (Tcb->CtrlFlag, TCP_CTRL_FIN_ACKED)) {
TcpSetState (Tcb, TCP_CLOSED);
}
break;
case TCP_TIME_WAIT:
TcpSendAck (Tcb);
if (Tcb->TimeWaitTimeout != 0) {
TcpSetTimer (Tcb, TCP_TIMER_2MSL, Tcb->TimeWaitTimeout);
} else {
DEBUG (
(EFI_D_WARN,
"Connection closed immediately because app disables TIME_WAIT timer for %p\n",
Tcb)
);
TcpClose (Tcb);
}
break;
default:
break;
}
//
// Sixth step: Check the URG bit.update the Urg point
// if in TCP_CAN_RECV, otherwise, leave the RcvUp intact.
//
StepSix:
Tcb->Idle = 0;
TcpSetKeepaliveTimer (Tcb);
if (TCP_FLG_ON (Seg->Flag, TCP_FLG_URG) && !TCP_FIN_RCVD (Tcb->State)) {
DEBUG (
(EFI_D_NET,
"TcpInput: received urgent data from peer for connected TCB %p\n",
Tcb)
);
Urg = Seg->Seq + Seg->Urg;
if (TCP_FLG_ON (Tcb->CtrlFlag, TCP_CTRL_RCVD_URG) &&
TCP_SEQ_GT (Urg, Tcb->RcvUp))
{
Tcb->RcvUp = Urg;
} else {
Tcb->RcvUp = Urg;
TCP_SET_FLG (Tcb->CtrlFlag, TCP_CTRL_RCVD_URG);
}
}
//
// Seventh step: Process the segment data
//
if (Seg->End != Seg->Seq) {
if (TCP_FIN_RCVD (Tcb->State)) {
DEBUG (
(EFI_D_WARN,
"TcpInput: connection reset because data is lost for connected TCB %p\n",
Tcb)
);
goto RESET_THEN_DROP;
}
if (TCP_LOCAL_CLOSED (Tcb->State) && (Nbuf->TotalSize != 0)) {
DEBUG (
(EFI_D_WARN,
"TcpInput: connection reset because data is lost for connected TCB %p\n",
Tcb)
);
goto RESET_THEN_DROP;
}
if (TcpQueueData (Tcb, Nbuf) == 0) {
DEBUG (
(EFI_D_ERROR,
"TcpInput: discard a broken segment for TCB %p\n",
Tcb)
);
goto DISCARD;
}
if (TcpDeliverData (Tcb) == -1) {
goto RESET_THEN_DROP;
}
if (!IsListEmpty (&Tcb->RcvQue)) {
TCP_SET_FLG (Tcb->CtrlFlag, TCP_CTRL_ACK_NOW);
}
}
//
// Eighth step: check the FIN.
// This step is moved to TcpDeliverData. FIN will be
// processed in sequence there. Check the comments in
// the beginning of the file header for information.
//
//
// Tcb is a new child of the listening Parent,
// commit it.
//
if (Parent != NULL) {
Tcb->Parent = Parent;
TcpInsertTcb (Tcb);
}
if ((Tcb->State != TCP_CLOSED) &&
(TcpToSendData (Tcb, 0) == 0) &&
(TCP_FLG_ON (Tcb->CtrlFlag, TCP_CTRL_ACK_NOW) || (Nbuf->TotalSize != 0)))
{
TcpToSendAck (Tcb);
}
NetbufFree (Nbuf);
return 0;
RESET_THEN_DROP:
TcpSendReset (Tcb, Head, Len, Dst, Src, Version);
DROP_CONNECTION:
ASSERT ((Tcb != NULL) && (Tcb->Sk != NULL));
NetbufFree (Nbuf);
TcpClose (Tcb);
return -1;
SEND_RESET:
TcpSendReset (Tcb, Head, Len, Dst, Src, Version);
DISCARD:
//
// Tcb is a child of Parent, and it doesn't survive
//
DEBUG ((EFI_D_WARN, "TcpInput: Discard a packet\n"));
NetbufFree (Nbuf);
if ((Parent != NULL) && (Tcb != NULL)) {
ASSERT (Tcb->Sk != NULL);
TcpClose (Tcb);
}
return 0;
}
/**
Process the received ICMP error messages for TCP.
@param[in] Nbuf The buffer that contains part of the TCP segment without an IP header
truncated from the ICMP error packet.
@param[in] IcmpErr The ICMP error code interpreted from an ICMP error packet.
@param[in] Src Source address of the ICMP error message.
@param[in] Dst Destination address of the ICMP error message.
@param[in] Version IP_VERSION_4 indicates IP4 stack. IP_VERSION_6 indicates
IP6 stack.
**/
VOID
TcpIcmpInput (
IN NET_BUF *Nbuf,
IN UINT8 IcmpErr,
IN EFI_IP_ADDRESS *Src,
IN EFI_IP_ADDRESS *Dst,
IN UINT8 Version
)
{
TCP_HEAD *Head;
TCP_CB *Tcb;
TCP_SEQNO Seq;
EFI_STATUS IcmpErrStatus;
BOOLEAN IcmpErrIsHard;
BOOLEAN IcmpErrNotify;
if (Nbuf->TotalSize < sizeof (TCP_HEAD)) {
goto CLEAN_EXIT;
}
Head = (TCP_HEAD *) NetbufGetByte (Nbuf, 0, NULL);
ASSERT (Head != NULL);
Tcb = TcpLocateTcb (
Head->DstPort,
Dst,
Head->SrcPort,
Src,
Version,
FALSE
);
if (Tcb == NULL || Tcb->State == TCP_CLOSED) {
goto CLEAN_EXIT;
}
//
// Validate the sequence number.
//
Seq = NTOHL (Head->Seq);
if (!(TCP_SEQ_LEQ (Tcb->SndUna, Seq) && TCP_SEQ_LT (Seq, Tcb->SndNxt))) {
goto CLEAN_EXIT;
}
IcmpErrStatus = IpIoGetIcmpErrStatus (
IcmpErr,
Tcb->Sk->IpVersion,
&IcmpErrIsHard,
&IcmpErrNotify
);
if (IcmpErrNotify) {
SOCK_ERROR (Tcb->Sk, IcmpErrStatus);
}
if (IcmpErrIsHard) {
TcpClose (Tcb);
}
CLEAN_EXIT:
NetbufFree (Nbuf);
}