kolibrios/kernel/trunk/network/tcp_output.inc

742 lines
21 KiB
PHP
Raw Normal View History

;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;; ;;
;; Copyright (C) KolibriOS team 2004-2017. 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_BIT_SENDALOT = 1 shl 0
;-----------------------------------------------------------------;
; ;
; tcp_output ;
; ;
; IN: eax = socket pointer ;
; ;
; OUT: eax = 0 on success/errorcode ;
; ;
;-----------------------------------------------------------------;
align 4
proc tcp_output
locals
temp_bits db ?
window dd ?
endl
DEBUGF DEBUG_NETWORK_VERBOSE, "TCP_output: socket=%x state=%u\n", eax, [eax + TCP_SOCKET.t_state]
push eax
lea ecx, [eax + SOCKET.mutex]
call mutex_lock
pop eax
DEBUGF DEBUG_NETWORK_VERBOSE, "TCP_output: socket locked\n"
; We'll detect the length of the data to be transmitted, and flags to be used
; If there is some data, or any critical controls to send (SYN / RST), then transmit
; Otherwise, investigate further
mov ebx, [eax + TCP_SOCKET.SND_MAX]
cmp ebx, [eax + TCP_SOCKET.SND_UNA]
jbe .not_idle
mov ebx, [eax + TCP_SOCKET.t_idle]
cmp ebx, [eax + TCP_SOCKET.t_rxtcur]
jbe .not_idle
; We have been idle for a while and no ACKS are expected to clock out any data we send..
; Slow start to get ack "clock" running again.
mov ebx, [eax + TCP_SOCKET.t_maxseg]
mov [eax + TCP_SOCKET.SND_CWND], ebx
.not_idle:
.again:
mov [temp_bits], 0
; Calculate offset
mov ebx, [eax + TCP_SOCKET.SND_NXT]
sub ebx, [eax + TCP_SOCKET.SND_UNA]
; Determine window
mov ecx, [eax + TCP_SOCKET.SND_WND]
cmp ecx, [eax + TCP_SOCKET.SND_CWND]
jb @f
mov ecx, [eax + TCP_SOCKET.SND_CWND]
@@:
; get flags in dl
call tcp_outflags
;------------------------
; data being forced out ?
; If in persist timeout with window of 0, send 1 byte.
; Otherwise, if window is small but nonzero, and timer expired,
; we will send what we can and go to transmit state
cmp [eax + TCP_SOCKET.t_force], 0
je .no_force
DEBUGF DEBUG_NETWORK_VERBOSE, "TCP_output: forcing data out\n"
test ecx, ecx
jnz .no_zero_window
cmp ebx, [eax + STREAM_SOCKET.snd.size]
jae @f
and dl, not (TH_FIN)
@@:
inc ecx
jmp .no_force
.no_zero_window:
and [eax + TCP_SOCKET.timer_flags], not timer_flag_persist
mov [eax + TCP_SOCKET.t_rxtshift], 0
.no_force:
;--------------------------------
; Calculate how much data to send
mov esi, [eax + STREAM_SOCKET.snd.size]
cmp esi, ecx
jb @f
mov esi, ecx
@@:
sub esi, ebx
;------------------------
; check for window shrink
; If FIN has been sent, but not ACKed, but we havent been called to retransmit, esi will be -1
; Otherwise, window shrank after we sent into it.
jae .not_persist
; enter persist state
xor esi, esi
; If window shrank to 0
test ecx, ecx
jnz @f
; cancel pending retransmit
and [eax + TCP_SOCKET.timer_flags], not timer_flag_retransmission
; pull SND_NXT back to (closed) window, We will enter persist state below.
push [eax + TCP_SOCKET.SND_UNA]
pop [eax + TCP_SOCKET.SND_NXT]
@@:
; If window didn't close completely, just wait for an ACK
.not_persist:
;---------------------------
; Send one segment at a time
cmp esi, [eax + TCP_SOCKET.t_maxseg]
jbe @f
mov esi, [eax + TCP_SOCKET.t_maxseg]
or [temp_bits], TCP_BIT_SENDALOT
@@:
;--------------------------------------------
; Turn of FIN flag if send buffer not emptied
mov edi, [eax + TCP_SOCKET.SND_NXT]
add edi, esi
sub edi, [eax + TCP_SOCKET.SND_UNA]
cmp edi, [eax + STREAM_SOCKET.snd.size]
jae @f
and dl, not (TH_FIN)
@@:
;-------------------------------
; calculate window advertisement
mov ecx, SOCKET_BUFFER_SIZE
sub ecx, [eax + STREAM_SOCKET.rcv.size]
;------------------------------
; Sender silly window avoidance
test esi, esi
jz .len_zero
cmp esi, [eax + TCP_SOCKET.t_maxseg]
je .send
add ebx, esi ; offset + length
cmp ebx, [eax + STREAM_SOCKET.snd.size]
jb @f
test [eax + TCP_SOCKET.t_flags], TF_NODELAY
jnz .send
mov ebx, [eax + TCP_SOCKET.SND_MAX]
cmp ebx, [eax + TCP_SOCKET.SND_UNA]
je .send
@@:
test [eax + TCP_SOCKET.t_force], -1 ;;;
jnz .send
mov ebx, [eax + TCP_SOCKET.max_sndwnd]
shr ebx, 1
cmp esi, ebx
jae .send
mov ebx, [eax + TCP_SOCKET.SND_NXT]
cmp ebx, [eax + TCP_SOCKET.SND_MAX]
jb .send
.len_zero:
;----------------------------------------
; Check if a window update should be sent
DEBUGF DEBUG_NETWORK_VERBOSE, "TCP_output: window=%d\n", ecx
; Compare available window to amount of window known to peer (as advertised window less next expected input)
; If the difference is at least two max size segments, or at least 50% of the maximum possible window,
; Then we want to send a window update to the peer.
test ecx, ecx
jz .no_window
push ecx
mov cl, [eax + TCP_SOCKET.RCV_SCALE]
mov ebx, TCP_max_win
shl ebx, cl
pop ecx
cmp ebx, ecx
jb @f
mov ebx, ecx
@@:
sub ebx, [eax + TCP_SOCKET.RCV_ADV]
add ebx, [eax + TCP_SOCKET.RCV_NXT]
DEBUGF DEBUG_NETWORK_VERBOSE, "TCP_output: we can increase window by %d bytes\n", ebx
mov edi, [eax + TCP_SOCKET.t_maxseg]
shl edi, 1
cmp ebx, edi
jae .send
shl ebx, 1
; cmp ebx, [eax + TCP_SOCKET.] ;;; TODO: check with receive buffer high water mark
; jae TCP_send
.no_window:
;--------------------------
; Should a segment be sent?
DEBUGF DEBUG_NETWORK_VERBOSE, "TCP_output: Should a segment be sent?\n"
test [eax + TCP_SOCKET.t_flags], TF_ACKNOW ; we need to ACK
jnz .send
test dl, TH_SYN + TH_RST ; we need to send a SYN or RST
jnz .send
mov ebx, [eax + TCP_SOCKET.SND_UP] ; when urgent pointer is beyond start of send bufer
cmp ebx, [eax + TCP_SOCKET.SND_UNA]
ja .send
; Do we need to send a FIN according to our state?
test dl, TH_FIN
jz .enter_persist ; no reason to send, enter persist state
; Do so if we didnt do it already
test [eax + TCP_SOCKET.t_flags], TF_SENTFIN
jz .send
; Or when we need to retransmit the FIN
mov ebx, [eax + TCP_SOCKET.SND_NXT]
cmp ebx, [eax + TCP_SOCKET.SND_UNA]
je .send
;--------------------
; Enter persist state
.enter_persist:
cmp [eax + STREAM_SOCKET.snd.size], 0 ; Data ready to send?
jne @f
and [eax + TCP_SOCKET.timer_flags], not timer_flag_retransmission
jne @f
test [eax + TCP_SOCKET.timer_flags], timer_flag_persist ; Persist timer already expired?
jnz @f
DEBUGF DEBUG_NETWORK_VERBOSE, "TCP_output: Entering persist state\n"
mov [eax + TCP_SOCKET.t_rxtshift], 0
call tcp_set_persist
@@:
;----------------------------
; No reason to send a segment
DEBUGF DEBUG_NETWORK_VERBOSE, "TCP_output: No reason to send a segment\n"
pusha
lea ecx, [eax + SOCKET.mutex]
call mutex_unlock
popa
xor eax, eax
ret
;-----------------------------------------------
;
; Send a segment
;
; eax = socket pointer
; esi = data len
; dl = flags
;
;-----------------------------------------------
.send:
DEBUGF DEBUG_NETWORK_VERBOSE, "TCP_send: socket=%x length=%u flags=%x\n", eax, esi, dl
push eax ; save socket ptr
push esi ; and data length too
mov edi, sizeof.TCP_header ; edi will contain headersize
;------------------------------------
; Send options with first SYN segment
test dl, TH_SYN
jz .options_done
push [eax + TCP_SOCKET.ISS]
pop [eax + TCP_SOCKET.SND_NXT]
test [eax + TCP_SOCKET.t_flags], TF_NOOPT
jnz .options_done
mov ecx, 1460 ;;;; FIXME: use routing blablabla to determine MSS
or ecx, TCP_OPT_MAXSEG shl 24 + 4 shl 16
bswap ecx
push ecx
add di, 4
DEBUGF DEBUG_NETWORK_VERBOSE, "TCP_send: added maxseg option\n"
test [eax + TCP_SOCKET.t_flags], TF_REQ_SCALE
jz .no_scale
test dl, TH_ACK
jz .scale_opt
test [eax + TCP_SOCKET.t_flags], TF_RCVD_SCALE
jz .no_scale
.scale_opt:
mov cl, [eax + TCP_SOCKET.request_r_scale]
mov ch, TCP_OPT_NOP
pushw cx
pushw TCP_OPT_WINDOW + 3 shl 8
add di, 4
DEBUGF DEBUG_NETWORK_VERBOSE, "TCP_send: added scale option\n"
.no_scale:
.no_syn:
;------------------------------------
; Make the timestamp option if needed
test [eax + TCP_SOCKET.t_flags], TF_REQ_TSTMP
jz .no_timestamp
test dl, TH_RST
jnz .no_timestamp
test dl, TH_ACK
jz .timestamp
test [eax + TCP_SOCKET.t_flags], TF_RCVD_TSTMP
jz .no_timestamp
.timestamp:
pushd 0
pushd [timer_ticks]
pushd TCP_OPT_NOP + TCP_OPT_NOP shl 8 + TCP_OPT_TIMESTAMP shl 16 + 10 shl 24
add di, 12
DEBUGF DEBUG_NETWORK_VERBOSE, "TCP_send: added timestamp\n"
.no_timestamp:
; <Add additional options here>
.options_done:
; eax = socket ptr
; edx = flags
; edi = header size
; esi = data len
;---------------------------------------------
; check if we dont exceed the max segment size
add esi, edi ; total TCP segment size
cmp esi, [eax + TCP_SOCKET.t_maxseg]
jbe .no_overflow
mov esi, [eax + TCP_SOCKET.t_maxseg]
or [temp_bits], TCP_BIT_SENDALOT
.no_overflow:
; Update stats
test esi, esi
jz .zero_data
cmp [eax + TCP_SOCKET.t_force], 1
jne @f
cmp esi, 1
jne @f
inc [TCPS_sndprobe]
jmp .eos
@@:
mov ebx, [eax + TCP_SOCKET.SND_NXT]
cmp ebx, [eax + TCP_SOCKET.SND_MAX]
jae @f
inc [TCPS_sndrexmitpack]
add [TCPS_sndrexmitbyte], esi
jmp .eos
@@:
inc [TCPS_sndpack]
add [TCPS_sndbyte], esi
jmp .eos
.zero_data:
test [eax + TCP_SOCKET.t_flags], TF_ACKNOW
jz @f
inc [TCPS_sndacks]
jmp .eos
@@:
test dl, TH_SYN + TH_FIN + TH_RST
jz @f
inc [TCPS_sndctrl]
jmp .eos
@@:
mov ebx, [eax + TCP_SOCKET.SND_UP]
cmp ebx, [eax + TCP_SOCKET.SND_UNA]
jb @f
inc [TCPS_sndurg]
jmp .eos
@@:
inc [TCPS_sndwinup]
.eos:
;---------------------------------------------------
; Dont increase sequence number when resending a FIN
test dl, TH_FIN
jz .no_fin_retransmit
test [eax + TCP_SOCKET.t_flags], TF_SENTFIN
jz .no_fin_retransmit
mov ebx, [eax + TCP_SOCKET.SND_NXT]
cmp ebx, [eax + TCP_SOCKET.SND_MAX]
jne .no_fin_retransmit
dec [eax + TCP_SOCKET.SND_NXT]
.no_fin_retransmit:
;----------------------------------------------------
; Calculate the receive window.
; Dont shrink window, but avoid silly window syndrome
mov ebx, SOCKET_BUFFER_SIZE
sub ebx, [eax + STREAM_SOCKET.rcv.size]
cmp ebx, SOCKET_BUFFER_SIZE/4
jae @f
cmp ebx, [eax + TCP_SOCKET.t_maxseg]
jae @f
xor ebx, ebx
@@:
cmp ebx, TCP_max_win
jbe @f
mov ebx, TCP_max_win
@@:
mov ecx, [eax + TCP_SOCKET.RCV_ADV]
sub ecx, [eax + TCP_SOCKET.RCV_NXT]
cmp ebx, ecx
jg @f
mov ebx, ecx
@@:
DEBUGF DEBUG_NETWORK_VERBOSE, "TCP_send: window=%u\n", ebx
mov [window], ebx
mov cl, [eax + TCP_SOCKET.RCV_SCALE]
shr ebx, cl
xchg bl, bh
;-----------------------------------------------------------------
; Start by pushing all TCP header values in reverse order on stack
; (essentially, creating the tcp header on the stack!)
pushw 0 ; UrgentPointer
pushw 0 ; Checksum
pushw bx ; Window
shl edi, 2 ; DataOffset
shl dx, 8
or dx, di ; Flags
pushw dx
shr edi, 2 ; DataOffset
push [eax + TCP_SOCKET.RCV_NXT] ; AckNumber
ntohd [esp]
push [eax + TCP_SOCKET.SND_NXT] ; SequenceNumber
ntohd [esp]
push [eax + TCP_SOCKET.RemotePort] ; DestinationPort
push [eax + TCP_SOCKET.LocalPort] ; SourcePort
push edi ; header size
;---------------------
; Create the IP packet
mov ecx, esi
mov ebx, [eax + IP_SOCKET.device]
mov edx, [eax + IP_SOCKET.LocalIP] ; source ip
mov edi, [eax + IP_SOCKET.RemoteIP] ; dest ip
mov al, [eax + IP_SOCKET.ttl]
mov ah, IP_PROTO_TCP
call ipv4_output
jz .ip_error
;------------------------------------------
; Move TCP header from stack to TCP segment
push ecx
mov ecx, [esp + 4]
lea esi, [esp + 8]
shr ecx, 2 ; count is in bytes, we will work with dwords
rep movsd
pop ecx ; full TCP packet size
pop esi ; headersize
add esp, esi ; remove it from stack
push eax ; packet ptr for send proc
mov edx, edi ; begin of data
sub edx, esi ; begin of packet (edi = begin of data)
push ecx
sub ecx, esi ; data size
;--------------
; Copy the data
; eax = ptr to ring struct
; ecx = buffer size
; edi = ptr to buffer
mov eax, [esp + 12] ; get socket ptr
push edx
push [eax + TCP_SOCKET.SND_NXT] ; we'll need this for timing the transmission
test ecx, ecx
jz .nodata
mov edx, [eax + TCP_SOCKET.SND_NXT]
add [eax + TCP_SOCKET.SND_NXT], ecx ; update sequence number
sub edx, [eax + TCP_SOCKET.SND_UNA] ; offset
add eax, STREAM_SOCKET.snd
call socket_ring_read
.nodata:
pop edi
pop esi ; begin of data
pop ecx ; full packet size
mov eax, [esp + 8] ; socket ptr
;----------------------------
; initialize retransmit timer
;TODO: check t_force and persist
test [esi + TCP_header.Flags], TH_SYN + TH_FIN ; syn and fin take a sequence number
jz @f
inc [eax + TCP_SOCKET.SND_NXT]
test [esi + TCP_header.Flags], TH_FIN
jz @f
or [eax + TCP_SOCKET.t_flags], TF_SENTFIN ; if we sent a fin, set the sentfin flag
@@:
mov edx, [eax + TCP_SOCKET.SND_NXT]
cmp edx, [eax + TCP_SOCKET.SND_MAX] ; is this a retransmission?
jbe @f
mov [eax + TCP_SOCKET.SND_MAX], edx ; [eax + TCP_SOCKET.SND_NXT] from before we updated it
cmp [eax + TCP_SOCKET.t_rtt], 0 ; are we currently timing anything?
je @f
mov [eax + TCP_SOCKET.t_rtt], 1 ; nope, start transmission timer
mov [eax + TCP_SOCKET.t_rtseq], edi
inc [TCPS_segstimed]
@@:
; set retransmission timer if not already set, and not doing an ACK or keepalive probe
test [eax + TCP_SOCKET.timer_flags], timer_flag_retransmission
jnz .retransmit_set
cmp edx, [eax + TCP_SOCKET.SND_UNA] ; edx is still [eax + TCP_SOCKET.SND_NXT]
je .retransmit_set
mov edx, [eax + TCP_SOCKET.t_rxtcur]
mov [eax + TCP_SOCKET.timer_retransmission], edx
or [eax + TCP_SOCKET.timer_flags], timer_flag_retransmission
test [eax + TCP_SOCKET.timer_flags], timer_flag_persist
jz .retransmit_set
and [eax + TCP_SOCKET.timer_flags], not timer_flag_persist
mov [eax + TCP_SOCKET.t_rxtshift], 0
.retransmit_set:
;--------------------
; Create the checksum
xor dx, dx
test [ebx + NET_DEVICE.hwacc], NET_HWACC_TCP_IPv4_OUT
jnz .checksum_ok
tcp_checksum (eax + IP_SOCKET.LocalIP), (eax + IP_SOCKET.RemoteIP)
.checksum_ok:
mov [esi + TCP_header.Checksum], dx
;----------------
; Send the packet
DEBUGF DEBUG_NETWORK_VERBOSE, "TCP_send: Sending with device %x\n", ebx
call [ebx + NET_DEVICE.transmit]
jnz .send_error
;---------------
; Ok, data sent!
pop ecx
pop eax
call net_ptr_to_num4
inc [TCP_segments_tx + edi]
inc [TCPS_sndtotal]
; update advertised receive window
mov ecx, [window]
test ecx, ecx
jz @f
add ecx, [eax + TCP_SOCKET.RCV_NXT]
cmp ecx, [eax + TCP_SOCKET.RCV_ADV]
jbe @f
mov [eax + TCP_SOCKET.RCV_ADV], ecx
@@:
; update last ack sent
push [eax + TCP_SOCKET.RCV_NXT]
pop [eax + TCP_SOCKET.last_ack_sent]
; clear the ACK flags
and [eax + TCP_SOCKET.t_flags], not (TF_ACKNOW + TF_DELACK)
;--------------
; unlock socket
DEBUGF DEBUG_NETWORK_VERBOSE, "TCP_send: unlocking socket 0x%x\n", eax
push eax
lea ecx, [eax + SOCKET.mutex]
call mutex_unlock
pop eax
;-----------------------------
; Check if we need more output
test [temp_bits], TCP_BIT_SENDALOT
jnz tcp_output.again
DEBUGF DEBUG_NETWORK_VERBOSE, "TCP_send: success!\n"
xor eax, eax
ret
.ip_error:
pop ecx
add esp, ecx
add esp, 4
pop eax
mov [eax + TCP_SOCKET.timer_retransmission], TCP_time_re_min
or [eax + TCP_SOCKET.timer_flags], timer_flag_retransmission
lea ecx, [eax + SOCKET.mutex]
call mutex_unlock
DEBUGF DEBUG_NETWORK_ERROR, "TCP_send: IP error\n"
or eax, -1
ret
.send_error:
add esp, 4
pop eax
lea ecx, [eax + SOCKET.mutex]
call mutex_unlock
DEBUGF DEBUG_NETWORK_ERROR, "TCP_send: sending failed\n"
or eax, -2
ret
endp