kolibrios/kernel/branches/net/network/tcp_input.inc

1587 lines
32 KiB
PHP
Raw Normal View History

;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;; ;;
;; Copyright (C) KolibriOS team 2004-2011. All rights reserved. ;;
;; Distributed under terms of the GNU General Public License ;;
;; ;;
;; Part of the tcp/ip network stack for KolibriOS ;;
;; ;;
;; Written by hidnplayr@kolibrios.org ;;
;; ;;
;; Based on the code of 4.4BSD ;;
;; ;;
;; GNU GENERAL PUBLIC LICENSE ;;
;; Version 2, June 1991 ;;
;; ;;
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
$Revision$
;-----------------------------------------------------------------
;
; TCP_input:
;
; IN: [esp] = ptr to buffer
; [esp+4] = buffer size
; ebx = ptr to device struct
; ecx = segment size
; edx = ptr to TCP segment
;
; esi = ipv4 source address
; edi = ipv4 dest address
;
; OUT: /
;
;-----------------------------------------------------------------
align 4
TCP_input:
DEBUGF 1,"TCP_input size=%u ", ecx
; Offset must be greater than or equal to the size of the standard TCP header (20) and less than or equal to the TCP length.
movzx eax, [edx + TCP_segment.DataOffset]
and eax, 0xf0
shr al, 2
DEBUGF 1,"headersize=%u\n", eax
cmp eax, 20
jl .drop_not_locked
;-------------------------------
; Now, re-calculate the checksum
push eax ecx edx
pushw [edx + TCP_segment.Checksum]
mov [edx + TCP_segment.Checksum], 0
push esi edi
mov esi, edx
TCP_checksum (esp), (esp+4)
pop esi edi ; yes, swap them (we dont need dest addr)
pop cx ; previous checksum
cmp cx, dx
pop edx ecx esi
jnz .drop_not_locked
DEBUGF 1,"Checksum is correct\n"
sub ecx, esi ; update packet size
jl .drop_not_locked
DEBUGF 1,"we got %u bytes of data\n", ecx
;-----------------------------------------------------------------------------------------
; Check if this packet has a timestamp option (We do it here so we can process it quickly)
cmp esi, 20 + 12 ; Timestamp option is 12 bytes
jl .no_timestamp
je .is_ok
cmp byte [edx + TCP_segment.Data + 12], TCP_OPT_EOL ; end of option list
jne .no_timestamp
.is_ok:
test [edx + TCP_segment.Flags], TH_SYN ; SYN flag must not be set
jnz .no_timestamp
cmp dword [edx + TCP_segment.Data], 0x0101080a ; Timestamp header
jne .no_timestamp
DEBUGF 1,"timestamp ok\n"
; TODO: Parse the option
; TODO: Set a Bit in the TCP to tell all options are parsed
.no_timestamp:
;-------------------------------------------
; Convert Big-endian values to little endian
ntohd [edx + TCP_segment.SequenceNumber]
ntohd [edx + TCP_segment.AckNumber]
ntohw [edx + TCP_segment.Window]
ntohw [edx + TCP_segment.UrgentPointer]
ntohw [edx + TCP_segment.SourcePort]
ntohw [edx + TCP_segment.DestinationPort]
;------------------------------------------------------------
; Next thing to do is find the TCPS (thus, the socket pointer)
; IP Packet TCP Destination Port = local Port
; (IP Packet SenderAddress = Remote IP) OR (Remote IP = 0)
; (IP Packet TCP Source Port = remote Port) OR (remote Port = 0)
mov ebx, net_sockets
.socket_loop:
mov ebx, [ebx + SOCKET.NextPtr]
or ebx, ebx
jz .drop_with_reset_not_locked
cmp [ebx + SOCKET.Domain], AF_INET4
jne .socket_loop
cmp [ebx + SOCKET.Protocol], IP_PROTO_TCP
jne .socket_loop
mov ax, [edx + TCP_segment.DestinationPort]
cmp [ebx + TCP_SOCKET.LocalPort], ax
jne .socket_loop
mov eax, [ebx + IP_SOCKET.RemoteIP]
cmp eax, edi ; edi is source ip from packet
je @f
test eax, eax
jnz .socket_loop
@@:
mov ax, [ebx + TCP_SOCKET.RemotePort]
cmp [edx + TCP_segment.SourcePort] , ax
je .found_socket
test ax, ax
jnz .socket_loop
.found_socket:
DEBUGF 1,"Socket ptr: %x\n", ebx
; ebx now contains the pointer to the socket
;----------------------------
; Check if socket isnt closed
cmp [ebx + TCP_SOCKET.t_state], TCPS_CLOSED
je .drop_not_locked
;----------------
; Lock the socket
cmp [ebx + SOCKET.lock], 0
jne .drop_not_locked ;;; HACK ! HACK ! dirty fucking HACK ! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
add ebx, SOCKET.lock
DEBUGF 1,"lock: %x\n", [ebx]
call wait_mutex
sub ebx, SOCKET.lock
DEBUGF 1,"Socket locked\n"
;---------------------------------------
; unscale the window into a 32 bit value
movzx eax, [edx + TCP_segment.Window]
push ecx
mov cl, [ebx + TCP_SOCKET.SND_SCALE]
shl eax, cl
mov dword [edx + TCP_segment.Window], eax ; word after window is checksum, we dont need checksum anymore
pop ecx
;-----------------------------------
; Is this socket a listening socket?
test [ebx + SOCKET.options], SO_ACCEPTCON
jz .no_listening_socket
DEBUGF 1,"Accepting new connection\n"
mov [ebx + SOCKET.lock], 0
call SOCKET_fork
test eax, eax
jz .drop
push [edx - IPv4_Packet.DataOrOptional + IPv4_Packet.DestinationAddress] ;;; FIXME
pop [eax + IP_SOCKET.LocalIP]
push [edx + TCP_segment.DestinationPort]
pop [eax + TCP_SOCKET.LocalPort]
mov [eax + TCP_SOCKET.t_state], TCPS_LISTEN
; mov [ebx + SOCKET.lock], 0
mov ebx, eax
jmp .LISTEN
.no_listening_socket:
;-------------------------------------
; Reset idle timer and keepalive timer
mov [ebx + TCP_SOCKET.t_idle], 0
mov [ebx + TCP_SOCKET.timer_keepalive], TCP_time_keep_interval
;--------------------
; Process TCP options
cmp esi, 20 ; esi is headersize
je .no_options
DEBUGF 1,"Segment has options\n"
cmp [ebx + TCP_SOCKET.t_state], TCPS_LISTEN ; no options when in listen state
jz .not_uni_xfer ; also no header prediction
lea edi, [edx + TCP_segment.Data]
lea eax, [edx + esi]
.opt_loop:
cmp edi, eax
jge .no_options
cmp byte [edi], TCP_OPT_EOL ; end of option list?
jz .no_options
cmp byte [edi], TCP_OPT_NOP ; nop ?
jz .opt_nop
cmp byte [edi], TCP_OPT_MAXSEG
je .opt_maxseg
cmp byte [edi], TCP_OPT_WINDOW
je .opt_window
cmp byte [edi], TCP_OPT_TIMESTAMP
je .opt_timestamp
jmp .no_options ; If we reach here, some unknown options were received, skip them all!
.opt_nop:
inc edi
jmp .opt_loop
.opt_maxseg:
cmp byte [edi+1], 4
jne .no_options ; error occured, ignore all options!
test [edx + TCP_segment.Flags], TH_SYN
jz @f
movzx eax, word[edi+2]
rol ax, 8
DEBUGF 1,"Maxseg: %u\n", ax
mov [ebx + TCP_SOCKET.t_maxseg], eax
@@:
add edi, 4
jmp .opt_loop
.opt_window:
cmp byte [edi+1], 3
jne .no_options
test [edx + TCP_segment.Flags], TH_SYN
jz @f
DEBUGF 1,"Got window option\n"
;;;;;
@@:
add edi, 3
jmp .opt_loop
.opt_timestamp:
cmp byte [edi+1], 10
jne .no_options
DEBUGF 1,"Got timestamp option\n"
;;;;;
add edi, 10
jmp .opt_loop
.no_options:
;-----------------------------------------------------------------------
; Time to do some header prediction (Original Principle by Van Jacobson)
; There are two common cases for an uni-directional data transfer.
;
; General rule: the packets has no control flags, is in-sequence,
; window width didnt change and we're not retransmitting.
;
; Second rules:
; - If the length is 0 and the ACK moved forward, we're the sender side of the transfer.
; In this case we'll free the ACK'ed data and notify higher levels that we have free space in buffer
;
; - If the length is not 0 and the ACK didn't move, we're the receiver side of the transfer.
; If the packets are in order (data queue is empty), add the data to the socket buffer and request a delayed ACK
cmp [ebx + TCP_SOCKET.t_state], TCPS_ESTABLISHED
jnz .not_uni_xfer
test [edx + TCP_segment.Flags], TH_SYN + TH_FIN + TH_RST + TH_URG
jnz .not_uni_xfer
test [edx + TCP_segment.Flags], TH_ACK
jz .not_uni_xfer
mov eax, [edx + TCP_segment.SequenceNumber]
cmp eax, [ebx + TCP_SOCKET.RCV_NXT]
jne .not_uni_xfer
mov eax, dword [edx + TCP_segment.Window]
cmp eax, [ebx + TCP_SOCKET.SND_WND]
jne .not_uni_xfer
mov eax, [ebx + TCP_SOCKET.SND_NXT]
cmp eax, [ebx + TCP_SOCKET.SND_MAX]
jne .not_uni_xfer
;---------------------------------------
; check if we are sender in the uni-xfer
; If the following 4 conditions are all true, this segment is a pure ACK.
;
; - The segment contains no data.
test ecx, ecx
jnz .not_sender
; - The congestion window is greater than or equal to the current send window.
; This test is true only if the window is fully open, that is, the connection is not in the middle of slow start or congestion avoidance.
mov eax, [ebx + TCP_SOCKET.SND_CWND]
cmp eax, [ebx + TCP_SOCKET.SND_WND]
jl .not_uni_xfer
; - The acknowledgment field in the segment is less than or equal to the maximum sequence number sent.
mov eax, [edx + TCP_segment.AckNumber]
cmp eax, [ebx + TCP_SOCKET.SND_MAX]
jg .not_uni_xfer
; - The acknowledgment field in the segment is greater than the largest unacknowledged sequence number.
sub eax, [ebx + TCP_SOCKET.SND_UNA]
jle .not_uni_xfer
DEBUGF 1,"Header prediction: we are sender\n"
;---------------------------------
; Packet is a pure ACK, process it
; Update RTT estimators
;;; TODO
; Delete acknowledged bytes from send buffer
pusha
mov ecx, eax
lea eax, [ebx + STREAM_SOCKET.snd]
call SOCKET_ring_free
popa
; update window pointers
mov eax, [edx + TCP_segment.AckNumber]
mov [ebx + TCP_SOCKET.SND_UNA], eax
; Stop retransmit timer
mov [ebx + TCP_SOCKET.timer_retransmission], 0
; Awaken waiting processes
mov [ebx + SOCKET.lock], 0
mov eax, ebx
call SOCKET_notify_owner
; Generate more output
call TCP_output
jmp .drop_not_locked
;-------------------------------------------------
; maybe we are the receiver in the uni-xfer then..
.not_sender:
; - The amount of data in the segment is greater than 0 (data count is in ecx)
; - The acknowledgment field equals the largest unacknowledged sequence number. This means no data is acknowledged by this segment.
mov eax, [edx + TCP_segment.AckNumber]
cmp eax, [ebx + TCP_SOCKET.SND_UNA]
jne .not_uni_xfer
; - The reassembly list of out-of-order segments for the connection is empty (seg_next equals tp).
;;; TODO
jnz .not_uni_xfer
; Complete processing of received data
DEBUGF 1,"header prediction: we are receiver\nreceiving %u bytes of data\n", ecx
add [ebx + TCP_SOCKET.RCV_NXT], ecx ; Update sequence number with number of bytes we have copied
add esi, edx
lea eax, [ebx + STREAM_SOCKET.rcv]
call SOCKET_ring_write ; Add the data to the socket buffer
mov eax, ebx
call SOCKET_notify_owner
or [ebx + TCP_SOCKET.t_flags], TF_DELACK ; Set delayed ack flag
jmp .drop
;--------------------------------------------------
; Header prediction failed, do it the slow way
.not_uni_xfer:
DEBUGF 1,"Header prediction failed\n"
; Calculate receive window size
;;;; TODO: 444
cmp [ebx + TCP_SOCKET.t_state], TCPS_LISTEN
je .LISTEN
cmp [ebx + TCP_SOCKET.t_state], TCPS_SYN_SENT
je .SYN_SENT
jmp .NOT_LISTEN_OR_SYN_SENT
;-------------
; Passive Open
align 4
.LISTEN:
DEBUGF 1,"TCP state: listen\n"
test [edx + TCP_segment.Flags], TH_RST ;;; TODO: kill new socket on error
jnz .drop
test [edx + TCP_segment.Flags], TH_ACK
jnz .drop_with_reset
test [edx + TCP_segment.Flags], TH_SYN
jz .drop
;;; TODO: check if it's a broadcast or multicast, and drop if so
push [edx - IPv4_Packet.DataOrOptional + IPv4_Packet.SourceAddress] ;;; FIXME
pop [ebx + IP_SOCKET.RemoteIP]
push [edx + TCP_segment.SourcePort]
pop [ebx + TCP_SOCKET.RemotePort]
push [edx + TCP_segment.SequenceNumber]
pop [ebx + TCP_SOCKET.IRS]
push [TCP_sequence_num] ;;;;;
add [TCP_sequence_num], 64000 / 2
pop [ebx + TCP_SOCKET.ISS]
push [ebx + TCP_SOCKET.ISS]
pop [ebx + TCP_SOCKET.SND_NXT]
TCP_sendseqinit ebx
TCP_rcvseqinit ebx
mov [ebx + TCP_SOCKET.t_state], TCPS_SYN_RECEIVED
mov [ebx + TCP_SOCKET.t_flags], TF_ACKNOW
mov [ebx + TCP_SOCKET.timer_keepalive], TCP_time_keep_interval ;;;; macro
lea eax, [ebx + STREAM_SOCKET.snd]
call SOCKET_ring_create
lea eax, [ebx + STREAM_SOCKET.rcv]
call SOCKET_ring_create
;;; call SOCKET_notify_owner
jmp .trim_then_step6
;------------
; Active Open
align 4
.SYN_SENT:
DEBUGF 1,"TCP state: syn_sent\n"
test [edx + TCP_segment.Flags], TH_ACK
jz @f
mov eax, [edx + TCP_segment.AckNumber]
cmp eax, [ebx + TCP_SOCKET.ISS]
jle .drop_with_reset
cmp eax, [ebx + TCP_SOCKET.SND_MAX]
jg .drop_with_reset
@@:
test [edx + TCP_segment.Flags], TH_RST
jz @f
test [edx + TCP_segment.Flags], TH_ACK
jz .drop
mov eax, ebx
mov ebx, ECONNREFUSED
call TCP_drop
jmp .drop
@@:
test [edx + TCP_segment.Flags], TH_SYN
jz .drop
; at this point, segment seems to be valid
test [edx + TCP_segment.Flags], TH_ACK
jz .no_syn_ack
; now, process received SYN in response to an active open
mov eax, [edx + TCP_segment.AckNumber]
mov [ebx + TCP_SOCKET.SND_UNA], eax
cmp eax, [ebx + TCP_SOCKET.SND_NXT]
jle @f
mov [ebx + TCP_SOCKET.SND_NXT], eax
@@:
.no_syn_ack:
mov [ebx + TCP_SOCKET.timer_retransmission], 0 ; disable retransmission
push [edx + TCP_segment.SequenceNumber]
pop [ebx + TCP_SOCKET.IRS]
TCP_rcvseqinit ebx
or [ebx + TCP_SOCKET.t_flags], TF_ACKNOW
mov eax, [ebx + TCP_SOCKET.SND_UNA]
cmp eax, [ebx + TCP_SOCKET.ISS]
jle .simultaneous_open
test [edx + TCP_segment.Flags], TH_ACK
jz .simultaneous_open
DEBUGF 1,"TCP: active open\n"
;;; TODO: update stats
; set socket state to connected
mov [ebx + SOCKET.state],1 ;;;; FIXME
mov [ebx + TCP_SOCKET.t_state], TCPS_ESTABLISHED
;;; TODO: check if we should scale the connection (567-572)
;;; TODO: update RTT estimators
jmp .trim_then_step6
.simultaneous_open:
DEBUGF 1,"TCP: simultaneous open\n"
; We have received a syn but no ACK, so we are having a simultaneous open..
mov [ebx + TCP_SOCKET.t_state], TCPS_SYN_RECEIVED
;-------------------------------------
; Common processing for receipt of SYN
.trim_then_step6:
inc [edx + TCP_segment.SequenceNumber]
;;; TODO: Drop any received data that follows receive window (590)
mov eax, [edx + TCP_segment.SequenceNumber]
mov [ebx + TCP_SOCKET.RCV_UP], eax
dec eax
mov [ebx + TCP_SOCKET.SND_WL1], eax
jmp .ack_processed
.NOT_LISTEN_OR_SYN_SENT:
DEBUGF 1,"Slow TCP input: not listen or syn_sent state\n"
;--------------------------------------------
; Protection Against Wrapped Sequence Numbers
; First, check if timestamp is present
;;;; TODO 602
; Then, check if at least some bytes of data are within window
;;;; TODO
;----------------------------
; trim any data not in window
; check for duplicate data at beginning of segment
mov eax, [ebx + TCP_SOCKET.RCV_NXT]
sub eax, [edx + TCP_segment.SequenceNumber]
jle .no_duplicate
DEBUGF 1,"Uh oh.. %u bytes of duplicate data!\n", eax
test [edx + TCP_segment.Flags], TH_SYN
jz .no_dup_syn
; remove duplicate syn
and [edx + TCP_segment.Flags], not (TH_SYN)
inc [edx + TCP_segment.SequenceNumber]
cmp [edx + TCP_segment.UrgentPointer], 1
jle @f
dec [edx + TCP_segment.UrgentPointer]
jmp .dup_syn
@@:
and [edx + TCP_segment.Flags], not (TH_URG)
.dup_syn:
dec eax
.no_dup_syn:
; eax holds number of bytes to drop
; Check for entire duplicate packet
cmp eax, ecx
jge .duplicate
DEBUGF 1,"Going to drop %u out of %u bytes\n", eax, ecx
;;; TODO: apply figure 28.30
; Check for duplicate FIN
test [edx + TCP_segment.Flags], TH_FIN
jz @f
inc ecx
cmp eax, ecx
dec ecx
jne @f
mov eax, ecx
and [edx + TCP_segment.Flags], not TH_FIN
or [ebx + TCP_SOCKET.t_flags], TF_ACKNOW
jmp .no_duplicate
@@:
; Handle the case when a bound socket connects to itself
; Allow packets with a SYN and an ACKto continue with the processing
;-------------------------------------
; Generate duplicate ACK if nescessary
; This code also handles simultaneous half-open or self-connects
test eax, eax
jnz .drop_after_ack
cmp [edx + TCP_segment.Flags], TH_ACK
jz .drop_after_ack
.duplicate:
DEBUGF 1,"Duplicate received\n"
;----------------------------------------
; Update statistics for duplicate packets
;;; TODO
jmp .drop_after_ack
.no_duplicate:
;-----------------------------------------------
; Remove duplicate data and update urgent offset
add [edx + TCP_segment.SequenceNumber], eax
;;; TODO
sub [edx + TCP_segment.UrgentPointer], ax
jg @f
and [edx + TCP_segment.Flags], not (TH_URG)
mov [edx + TCP_segment.UrgentPointer], 0
@@:
;--------------------------------------------------
; Handle data that arrives after process terminates
cmp [ebx + SOCKET.PID], 0
jg @f
cmp [ebx + TCP_SOCKET.t_state], TCPS_CLOSE_WAIT
jle @f
test ecx, ecx
jz @f
;;; Close the socket
;;; update stats
jmp .drop_with_reset
@@:
;----------------------------------------
; Remove data beyond right edge of window
mov eax, [edx + TCP_segment.SequenceNumber]
add eax, ecx
sub eax, [ebx + TCP_SOCKET.RCV_NXT]
sub ax, [ebx + TCP_SOCKET.RCV_WND]
; eax now holds the number of bytes to drop
jle .no_excess_data
;;; TODO: update stats
cmp eax, ecx
jl .dont_drop_all
;;; TODO 700-736
.dont_drop_all:
.no_excess_data:
;-----------------
; Record timestamp
;;; TODO 737-746
;------------------
; Process RST flags
test [edx + TCP_segment.Flags], TH_RST
jz .rst_skip
DEBUGF 1,"Got an RST flag"
mov eax, [ebx + TCP_SOCKET.t_state]
shl eax, 2
jmp dword [eax + .rst_sw_list]
.rst_sw_list:
dd .rst_skip ;TCPS_CLOSED
dd .rst_skip ;TCPS_LISTEN
dd .rst_skip ;TCPS_SYN_SENT
dd .econnrefused ;TCPS_SYN_RECEIVED
dd .econnreset ;TCPS_ESTABLISHED
dd .econnreset ;TCPS_CLOSE_WAIT
dd .econnreset ;TCPS_FIN_WAIT_1
dd .rst_close ;TCPS_CLOSING
dd .rst_close ;TCPS_LAST_ACK
dd .econnreset ;TCPS_FIN_WAIT_2
dd .rst_close ;TCPS_TIMED_WAIT
.econnrefused:
DEBUGF 1,"Connection refused"
mov [ebx + SOCKET.errorcode], ECONNREFUSED
jmp .close
.econnreset:
DEBUGF 1,"Connection reset"
mov [ebx + SOCKET.errorcode], ECONNRESET
.close:
DEBUGF 1,"Closing connection"
mov [ebx + TCP_SOCKET.t_state], TCPS_CLOSED
;;; TODO: update stats
mov eax, ebx
call TCP_close
jmp .drop
.rst_close:
DEBUGF 1,"Closing with reset\n"
mov eax, ebx
call TCP_close
jmp .drop
.rst_skip:
;--------------------------------------
; handle SYN-full and ACK-less segments
test [edx + TCP_segment.Flags], TH_SYN
jz @f
mov eax, ebx
mov ebx, ECONNRESET
call TCP_drop
jmp .drop_with_reset
test [edx + TCP_segment.Flags], TH_ACK
jz .drop
@@:
;---------------
; ACK processing
cmp [ebx + TCP_SOCKET.t_state], TCPS_SYN_RECEIVED
jnz .no_syn_rcv
DEBUGF 1,"TCP state = syn received\n"
mov eax, [edx + TCP_segment.AckNumber]
cmp [ebx + TCP_SOCKET.SND_UNA], eax
jg .drop_with_reset
cmp eax, [ebx + TCP_SOCKET.SND_MAX]
jg .drop_with_reset
;;; update stats
mov eax, ebx
call SOCKET_is_connected
mov [ebx + TCP_SOCKET.t_state], TCPS_ESTABLISHED
; Do window scaling?
test [ebx + TCP_SOCKET.t_flags], TF_RCVD_SCALE
jz @f
test [ebx + TCP_SOCKET.t_flags], TF_REQ_SCALE
jz @f
;;; 810-811
@@:
;;; 813 ?
mov eax, [edx + TCP_segment.SequenceNumber]
dec eax
mov [ebx + TCP_SOCKET.SND_WL1], eax
jmp .not_dup_ack
.no_syn_rcv:
; check for duplicate ACK
mov eax, [edx + TCP_segment.AckNumber]
cmp eax, [ebx + TCP_SOCKET.SND_UNA]
jg .not_dup_ack
test ecx, ecx
jnz .reset_dupacks
mov eax, dword [edx + TCP_segment.Window]
cmp eax, [ebx + TCP_SOCKET.SND_WND]
jne .reset_dupacks
DEBUGF 1,"Processing a duplicate ACK..\n"
cmp [ebx + TCP_SOCKET.timer_retransmission], 10000 ;;;;
jg @f
mov eax, [edx + TCP_segment.AckNumber]
cmp eax, [ebx + TCP_SOCKET.SND_UNA]
je .dup_ack
@@:
mov [ebx + TCP_SOCKET.t_dupacks], 0
jmp .not_dup_ack
.dup_ack:
inc [ebx + TCP_SOCKET.t_dupacks]
cmp [ebx + TCP_SOCKET.t_dupacks], TCP_re_xmit_thresh
jne .no_re_xmit
push [ebx + TCP_SOCKET.SND_NXT] ; >>>>
mov eax, [ebx + TCP_SOCKET.SND_WND]
cmp eax, [ebx + TCP_SOCKET.SND_CWND]
cmovg eax, [ebx + TCP_SOCKET.SND_CWND]
shr eax, 1
push edx
xor edx, edx
div [ebx + TCP_SOCKET.t_maxseg]
cmp eax, 2
jge @f
mov ax, 2
@@:
mul [ebx + TCP_SOCKET.t_maxseg]
pop edx
mov [ebx + TCP_SOCKET.SND_SSTHRESH], eax
mov [ebx + TCP_SOCKET.timer_retransmission], 0 ; turn off retransmission timer
mov [ebx + TCP_SOCKET.t_rtt], 0
mov eax, [edx + TCP_segment.AckNumber]
mov [ebx + TCP_SOCKET.SND_NXT], eax
mov eax, [ebx + TCP_SOCKET.t_maxseg]
mov [ebx + TCP_SOCKET.SND_CWND], eax
mov eax, ebx
call TCP_output ; retransmit missing segment
push edx
xor edx, edx
mov eax, [ebx + TCP_SOCKET.t_maxseg]
mul [ebx + TCP_SOCKET.t_dupacks]
pop edx
add eax, [ebx + TCP_SOCKET.SND_SSTHRESH]
mov [ebx + TCP_SOCKET.SND_CWND], eax
pop eax ; <<<<
cmp eax, [ebx + TCP_SOCKET.SND_NXT]
jl @f
mov [ebx + TCP_SOCKET.SND_NXT], eax
@@:
jmp .drop
.no_re_xmit:
jle .not_dup_ack
DEBUGF 1,"Increasing congestion window\n"
mov eax, [ebx + TCP_SOCKET.t_maxseg]
add [ebx + TCP_SOCKET.SND_CWND], eax
mov eax, ebx
call TCP_output
jmp .drop
.not_dup_ack:
;-------------------------------------------------
; If the congestion window was inflated to account
; for the other side's cached packets, retract it
mov eax, [ebx + TCP_SOCKET.SND_SSTHRESH]
cmp eax, [ebx + TCP_SOCKET.SND_CWND]
jg @f
cmp [ebx + TCP_SOCKET.t_dupacks], TCP_re_xmit_thresh
jle @f
mov [ebx + TCP_SOCKET.SND_CWND], eax
@@:
mov [ebx + TCP_SOCKET.t_dupacks], 0
mov eax, [edx + TCP_segment.AckNumber]
cmp eax, [ebx + TCP_SOCKET.SND_MAX]
jle @f
;;; TODO: update stats
jmp .drop_after_ack
@@:
mov edi, [edx + TCP_segment.AckNumber]
sub edi, [ebx + TCP_SOCKET.SND_UNA] ; now we got the number of acked bytes in esi
;;; TODO: update stats
DEBUGF 1,"We have an acceptable ACK of %x bytes\n", esi
;------------------------------------------
; RTT measurements and retransmission timer
;;;;; 912 - 926
mov [ebx + TCP_SOCKET.timer_retransmission], 0
mov eax, [ebx + TCP_SOCKET.SND_MAX]
cmp eax, [edx + TCP_segment.AckNumber]
je .all_outstanding
mov [ebx + TCP_SOCKET.timer_retransmission], 120 ;;;; TODO: correct this value (use a macro for it)
.all_outstanding:
;-------------------------------------------
; Open congestion window in response to ACKs
mov esi, [ebx + TCP_SOCKET.SND_CWND]
mov eax, [ebx + TCP_SOCKET.t_maxseg]
cmp esi, [ebx + TCP_SOCKET.SND_SSTHRESH]
jle @f
push edx
push eax
mul eax
div esi
pop edx
shr edx, 3
add eax, edx
pop edx
@@:
add esi, eax
push ecx
mov cl, [ebx + TCP_SOCKET.SND_SCALE]
mov eax, TCP_max_win
shl eax, cl
pop ecx
cmp esi, eax
cmovg esi, eax
mov [ebx + TCP_SOCKET.SND_CWND], esi
;------------------------------------------
; Remove acknowledged data from send buffer
push ecx edx ebx
mov ecx, edi
lea eax, [ebx + STREAM_SOCKET.snd]
call SOCKET_ring_free
pop ebx
sub [ebx + TCP_SOCKET.SND_WND], ecx
pop edx ecx
; Wake up process waiting on send buffer
mov eax, ebx
call SOCKET_notify_owner
; Update TCPS
mov eax, [edx + TCP_segment.AckNumber]
mov [ebx + TCP_SOCKET.SND_UNA], eax
cmp eax, [ebx + TCP_SOCKET.SND_NXT]
jl @f
mov [ebx + TCP_SOCKET.SND_NXT], eax
@@:
;; TODO: use zero flag as 'ourfinisacked'
; General ACK handling complete
; Now do the state-specific ones
mov eax, [ebx + TCP_SOCKET.t_state]
jmp dword [eax*4 + .ACK_sw_list]
.ACK_sw_list:
dd .ack_processed ;TCPS_CLOSED
dd .ack_processed ;TCPS_LISTEN
dd .ack_processed ;TCPS_SYN_SENT
dd .ack_processed ;TCPS_SYN_RECEIVED
dd .ack_processed ;TCPS_ESTABLISHED
dd .ack_processed ;TCPS_CLOSE_WAIT
dd .ack_fw1 ;TCPS_FIN_WAIT_1
dd .ack_c ;TCPS_CLOSING
dd .ack_la ;TCPS_LAST_ACK
dd .ack_processed ;TCPS_FIN_WAIT_2
dd .ack_tw ;TCPS_TIMED_WAIT
.ack_fw1:
jz .ack_processed
test [ebx + SOCKET.state], SS_CANTRCVMORE
jnz @f
mov eax, ebx
call SOCKET_is_disconnected
;;; mov [ebx + TCP_SOCKET.timer_timed_wait], TCP_time_max_idle
@@:
mov [ebx + TCP_SOCKET.t_state], TCPS_FIN_WAIT_2
jmp .ack_processed
.ack_c:
jz .ack_processed
mov [ebx + TCP_SOCKET.t_state], TCPS_TIMED_WAIT
mov eax, ebx
call TCP_cancel_timers
mov [ebx + TCP_SOCKET.timer_timed_wait], 2 * TCP_time_MSL
mov eax, ebx
call SOCKET_is_disconnected
jmp .ack_processed
.ack_la:
jz .ack_processed
mov eax, ebx
call TCP_close
jmp .drop
.ack_tw:
mov [ebx + TCP_SOCKET.timer_timed_wait], 2 * TCP_time_MSL
jmp .drop_after_ack
.reset_dupacks: ; We got a new ACK, reset duplicate ACK counter
mov [ebx + TCP_SOCKET.t_dupacks], 0
.ack_processed: ; (step 6)
DEBUGF 1,"ACK processed\n"
;----------------------------------------------
; check if we need to update window information
test [edx + TCP_segment.Flags], TH_ACK
jz .no_window_update
mov eax, [ebx + TCP_SOCKET.SND_WL1]
cmp eax, [edx + TCP_segment.SequenceNumber]
jl .update_window
jg @f
mov eax, [ebx + TCP_SOCKET.SND_WL2]
cmp eax, [edx + TCP_segment.AckNumber]
jl .update_window
jg .no_window_update
@@:
mov eax, [ebx + TCP_SOCKET.SND_WL2]
cmp eax, [edx + TCP_segment.AckNumber]
jne .no_window_update
movzx eax, [edx + TCP_segment.Window]
cmp eax, [ebx + TCP_SOCKET.SND_WND]
jle .no_window_update
.update_window:
DEBUGF 1,"Updating window\n"
; Keep track of pure window updates
; test ecx, ecx
; jz @f
;
; mov eax, [ebx + TCP_SOCKET.SND_WL2]
; cmp eax, [edx + TCP_segment.AckNumber]
; jne @f
;
; ;; mov eax, tiwin
; cmp eax, [ebx + TCP_SOCKET.SND_WND]
; jle @f
;
; ;;; update stats
;
; @@:
mov eax, dword [edx + TCP_segment.Window]
cmp eax, [ebx + TCP_SOCKET.max_sndwnd]
jle @f
mov [ebx + TCP_SOCKET.max_sndwnd], eax
@@:
mov [ebx + TCP_SOCKET.SND_WND], eax
push [edx + TCP_segment.SequenceNumber]
pop [ebx + TCP_SOCKET.SND_WL1]
push [edx + TCP_segment.AckNumber]
pop [ebx + TCP_SOCKET.SND_WL2]
;;; needoutput = 1
.no_window_update:
;-----------------
; process URG flag
test [edx + TCP_segment.Flags], TH_URG
jz .not_urgent
cmp [edx + TCP_segment.UrgentPointer], 0
jz .not_urgent
cmp [ebx + TCP_SOCKET.t_state], TCPS_TIMED_WAIT
je .not_urgent
; Ignore bogus urgent offsets
;;; 1040-1050
movzx eax, [edx + TCP_segment.UrgentPointer]
add eax, [ebx + STREAM_SOCKET.rcv + RING_BUFFER.size]
cmp eax, SOCKET_MAXDATA
jle .not_urgent
mov [edx + TCP_segment.UrgentPointer], 0
and [edx + TCP_segment.Flags], not (TH_URG)
jmp .do_data
.not_urgent:
; processing of received urgent pointer
;;; TODO (1051-1093)
;--------------------------------
; process the data in the segment
.do_data:
DEBUGF 1,"TCP: do data (%u)\n", ecx
test [edx + TCP_segment.Flags], TH_FIN
jnz .process_fin
cmp [ebx + TCP_SOCKET.t_state], TCPS_FIN_WAIT_1
jge .dont_do_data
test ecx, ecx
jz .final_processing
DEBUGF 1,"Processing data in segment\n"
;; TODO: check if data is in sequence !
movzx eax, [edx + TCP_segment.DataOffset] ;;; todo: remember this in.. edi ?
and eax, 0xf0
shr al, 2
lea esi, [edx + eax]
or [ebx + TCP_SOCKET.t_flags], TF_DELACK
add [ebx + TCP_SOCKET.RCV_NXT], ecx
lea eax, [ebx + STREAM_SOCKET.rcv]
call SOCKET_ring_write
mov eax, ebx
call SOCKET_notify_owner
jmp .final_processing
.dont_do_data:
;---------------
; FIN processing
.process_fin:
DEBUGF 1,"Processing FIN\n"
cmp [ebx + TCP_SOCKET.t_state], TCPS_CLOSE_WAIT
je .not_first_fin
cmp [ebx + TCP_SOCKET.t_state], TCPS_CLOSING
je .not_first_fin
cmp [ebx + TCP_SOCKET.t_state], TCPS_FIN_WAIT_2
je .not_first_fin
DEBUGF 1,"First FIN for this connection\n"
mov eax, ebx
call SOCKET_cant_recv_more
mov [ebx + TCP_SOCKET.t_flags], TF_ACKNOW
inc [ebx + TCP_SOCKET.RCV_NXT]
.not_first_fin:
mov eax, [ebx + TCP_SOCKET.t_state]
shl eax, 2
jmp dword [eax + .FIN_sw_list]
.FIN_sw_list:
dd .no_fin ;TCPS_CLOSED
dd .no_fin ;TCPS_LISTEN
dd .no_fin ;TCPS_SYN_SENT
dd .fin_syn_est ;TCPS_SYN_RECEIVED
dd .fin_syn_est ;TCPS_ESTABLISHED
dd .no_fin ;TCPS_CLOSE_WAIT
dd .fin_wait1 ;TCPS_FIN_WAIT_1
dd .no_fin ;TCPS_CLOSING
dd .no_fin ;TCPS_LAST_ACK
dd .fin_wait2 ;TCPS_FIN_WAIT_2
dd .fin_timed ;TCPS_TIMED_WAIT
.fin_syn_est:
mov [ebx + TCP_SOCKET.t_state], TCPS_CLOSE_WAIT
jmp .no_fin
.fin_wait1:
mov [ebx + TCP_SOCKET.t_state], TCPS_CLOSING
jmp .no_fin
.fin_wait2:
mov [ebx + TCP_SOCKET.t_state], TCPS_TIMED_WAIT
mov eax, ebx
call TCP_cancel_timers
mov [ebx + TCP_SOCKET.timer_timed_wait], 2 * TCP_time_MSL
call SOCKET_is_disconnected
jmp .no_fin
.fin_timed:
mov [ebx + TCP_SOCKET.timer_timed_wait], 2 * TCP_time_MSL
jmp .no_fin
.no_fin:
;-----------------
; Final processing
.final_processing:
DEBUGF 1,"Final processing\n"
;;; if debug enabled, output packet
;test needoutput, needoutput
;jz .dumpit
test [ebx + TCP_SOCKET.t_flags], TF_ACKNOW
jz .dumpit
DEBUGF 1,"ACK now!\n"
push ebx
mov eax, ebx
call TCP_output
pop ebx
.dumpit:
mov [ebx + SOCKET.lock], 0
call kernel_free
add esp, 4
ret
;------------------------------------------
; Generate an ACK, droping incoming segment
align 4
.drop_after_ack:
DEBUGF 1,"Drop after ACK\n"
test [edx + TCP_segment.Flags], TH_RST
jnz .drop
and [ebx + TCP_SOCKET.t_flags], TF_ACKNOW
mov [ebx + SOCKET.lock], 0
push ebx
; mov cl, TH_ACK
; call TCP_respond_socket
mov eax, ebx
call TCP_output
pop ebx
call kernel_free
add esp, 4
ret
;-------------------------------------------
; Generate an RST, dropping incoming segment
align 4
.drop_with_reset:
mov [ebx + SOCKET.lock], 0
.drop_with_reset_not_locked:
DEBUGF 1,"Drop with reset\n"
test [edx + TCP_segment.Flags], TH_RST
jnz .drop
;;; if its a multicast/broadcast, also drop
test [edx + TCP_segment.Flags], TH_ACK
jnz .respond_ack
test [edx + TCP_segment.Flags], TH_SYN
jnz .respond_syn
call kernel_free
add esp, 4
ret
.respond_ack:
push ebx
mov cl, TH_RST
call TCP_respond_socket
pop ebx
jmp .destroy_new_socket
.respond_syn:
push ebx
mov cl, TH_RST + TH_ACK
call TCP_respond_socket
pop ebx
jmp .destroy_new_socket
;-----
; Drop
align 4
.drop:
mov [ebx + SOCKET.lock], 0
.drop_not_locked:
DEBUGF 1,"Dropping packet\n"
;;;; If debugging options are enabled, output the packet somwhere
.destroy_new_socket:
;;;; kill the newly created socket
call kernel_free
add esp, 4
ret