Subversion Repositories Kolibri OS

;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;

;;                                                                 ;;

;; Copyright (C) KolibriOS team 2004-2015. 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 algorithms used in 4.4BSD                       ;;

;;                                                                 ;;

;;          GNU GENERAL PUBLIC LICENSE                             ;;

;;             Version 2, June 1991                                ;;

;;                                                                 ;;

;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;

$Revision: 6011 $

;-----------------------------------------------------------------;

;                                                                 ;

; TCP_input: Add a segment to the incoming TCP queue.             ;

;                                                                 ;

;  IN:  [esp] = ptr to buffer                                     ;

;       ebx = ptr to device struct                                ;

;       ecx = TCP segment size                                    ;

;       edx = ptr to IPv4 header                                  ;

;       esi = ptr to TCP segment                                  ;

;       edi = interface number*4                                  ;

;                                                                 ;

;  OUT: /                                                         ;

;                                                                 ;

;-----------------------------------------------------------------;

align 4

tcp_input:

; record the current time

        push    [timer_ticks]           ; in 1/100 seconds

        push    ebx ecx esi edx         ; mind the order (see TCP_queue_entry struct)

        mov     esi, esp

        push    edi

        add_to_queue TCP_queue, TCP_QUEUE_SIZE, sizeof.TCP_queue_entry, .fail

        pop     edi

        add     esp, sizeof.TCP_queue_entry

        inc     [TCP_segments_rx + edi]

        xor     edx, edx

        mov     eax, [TCP_input_event]

        mov     ebx, [eax + EVENT.id]

        xor     esi, esi

        call    raise_event

        ret

  .fail:

        pop     edi

        DEBUGF  DEBUG_NETWORK_VERBOSE, "TCP incoming queue is full, discarding packet!\n"

        call    net_ptr_to_num4

        inc     [TCP_segments_missed + edi]

        add     esp, sizeof.TCP_queue_entry - 4

        call    net_buff_free

        ret

align 4

proc tcp_process_input

locals

        dataoffset      dd ?

        timestamp       dd ?

        temp_bits       db ?

endl

        xor     esi, esi

        mov     ecx, MANUAL_DESTROY

        call    create_event

        mov     [TCP_input_event], eax

  .wait:

        mov     eax, [TCP_input_event]

        mov     ebx, [eax + EVENT.id]

        call    wait_event

  .loop:

        get_from_queue TCP_queue, TCP_QUEUE_SIZE, sizeof.TCP_queue_entry, .wait

        push    [esi + TCP_queue_entry.timestamp]

        pop     [timestamp]

        push    [esi + TCP_queue_entry.buffer_ptr]

        mov     ebx, [esi + TCP_queue_entry.device_ptr]

        mov     ecx, [esi + TCP_queue_entry.segment_size]

        mov     edi, [esi + TCP_queue_entry.ip_ptr]                     ; ptr to ipv4 header

        mov     esi, [esi + TCP_queue_entry.segment_ptr]                ; change esi last

        DEBUGF  DEBUG_NETWORK_VERBOSE, "TCP_input: size=%u time=%d\n", ecx, [timer_ticks]

        mov     edx, esi

        cmp     ebx, LOOPBACK_DEVICE

        je      .checksum_ok

; re-calculate the checksum (if not already done by hw)

        test    [ebx + NET_DEVICE.hwacc], NET_HWACC_TCP_IPv4_IN

        jnz     .checksum_ok

        push    ecx esi

        pushw   [esi + TCP_header.Checksum]

        mov     [esi + TCP_header.Checksum], 0

        tcp_checksum (edi+IPv4_header.SourceAddress), (edi+IPv4_header.DestinationAddress)

        pop     cx                              ; previous checksum

        cmp     cx, dx

        pop     edx ecx

        jne     .drop_no_socket

  .checksum_ok:

; Verify the data offset

        movzx   eax, [edx + TCP_header.DataOffset]

        and     al, 0xf0                        ; Calculate TCP segment header size (throwing away unused reserved bits in TCP header)

        shr     al, 2

        cmp     al, sizeof.TCP_header           ; Now see if it's at least the size of a standard TCP header

        jb      .drop_no_socket                 ; If not, drop the packet

        mov     [dataoffset], eax

        sub     ecx, eax                                                ; substract TCP header size from total segment size

        jb      .drop_no_socket                                         ; If total segment size is less then the advertised header size, drop packet

        DEBUGF  DEBUG_NETWORK_VERBOSE, "TCP_input: %u bytes of data\n", ecx

;-------------------------------------------

; Convert Big-endian values to little endian

        ntohd   [edx + TCP_header.SequenceNumber]

        ntohd   [edx + TCP_header.AckNumber]

        ntohw   [edx + TCP_header.Window]

        ntohw   [edx + TCP_header.UrgentPointer]

;------------------------

; Find 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)

  .findpcb:

        pusha

        mov     ecx, socket_mutex

        call    mutex_lock

        popa

        mov     ebx, net_sockets

        mov     si, [edx + TCP_header.DestinationPort]

  .socket_loop:

        mov     ebx, [ebx + SOCKET.NextPtr]

        or      ebx, ebx

        jz      .no_socket ;respond_seg_reset

        cmp     [ebx + SOCKET.Domain], AF_INET4

        jne     .socket_loop

        cmp     [ebx + SOCKET.Protocol], IP_PROTO_TCP

        jne     .socket_loop

        cmp     [ebx + TCP_SOCKET.LocalPort], si

        jne     .socket_loop

        mov     eax, [ebx + IP_SOCKET.RemoteIP]

        cmp     eax, [edi + IPv4_header.SourceAddress]

        je      @f

        test    eax, eax

        jnz     .socket_loop

       @@:

        mov     ax, [ebx + TCP_SOCKET.RemotePort]

        cmp     [edx + TCP_header.SourcePort], ax

        je      .found_socket

        test    ax, ax

        jnz     .socket_loop

  .found_socket:                                        ; ebx now contains the socketpointer

        pusha

        mov     ecx, socket_mutex

        call    mutex_unlock

        popa

        DEBUGF  DEBUG_NETWORK_VERBOSE, "TCP_input: socket ptr=%x state=%u flags=%x\n", ebx, [ebx + TCP_SOCKET.t_state], [edx + TCP_header.Flags]:2

;----------------------------

; Check if socket isnt closed

        cmp     [ebx + TCP_SOCKET.t_state], TCPS_CLOSED

        je      .drop_no_socket

;----------------

; Lock the socket

        pusha

        lea     ecx, [ebx + SOCKET.mutex]

        call    mutex_lock

        popa

        DEBUGF  DEBUG_NETWORK_VERBOSE, "TCP_input: socket locked\n"

;---------------------------

; disable all temporary bits

        mov     [temp_bits], 0

;---------------------------------------

; unscale the window into a 32 bit value

        movzx   eax, [edx + TCP_header.Window]

        push    ecx

        mov     cl, [ebx + TCP_SOCKET.SND_SCALE]

        shl     eax, cl

        mov     dword [edx + TCP_header.Window], eax    ; word after window is checksum, we dont need checksum anymore

        pop     ecx

;---------------------------------------

; Are we accepting incoming connections?

        test    [ebx + SOCKET.options], SO_ACCEPTCON

        jz      .no_accept

        DEBUGF  DEBUG_NETWORK_VERBOSE, "TCP_input: Accepting new connection\n"

        pusha

        lea     ecx, [ebx + SOCKET.mutex]

        call    mutex_unlock

        popa

        push    ecx edx esi edi

        call    socket_fork

        pop     edi esi edx ecx

        test    eax, eax

        jz      .drop_no_socket

        mov     ebx, eax

        mov     [temp_bits], TCP_BIT_DROPSOCKET

        push    [edi + IPv4_header.DestinationAddress]

        pop     [ebx + IP_SOCKET.LocalIP]

        push    [edx + TCP_header.DestinationPort]

        pop     [ebx + TCP_SOCKET.LocalPort]

        mov     [ebx + TCP_SOCKET.t_state], TCPS_LISTEN

  .no_accept:

;-------------------------------------

; Reset idle timer and keepalive timer

        mov     [ebx + TCP_SOCKET.t_idle], 0

        mov     [ebx + TCP_SOCKET.timer_keepalive], TCP_time_keep_idle

        or      [ebx + TCP_SOCKET.timer_flags], timer_flag_keepalive

;--------------------

; Process TCP options

;;; FIXME: for LISTEN, options should be called after we determined route, we need it for MSS

;;;        cmp     [ebx + TCP_SOCKET.t_state], TCPS_LISTEN ; no options when in listen state

;;;        jz      .not_uni_xfer                           ; also no header prediction

        push    ecx

        mov     ecx, [dataoffset]

        cmp     ecx, sizeof.TCP_header                  ; Does header contain any options?

        je      .no_options

        DEBUGF  DEBUG_NETWORK_VERBOSE, "TCP_input: Segment has options\n"

        add     ecx, edx

        lea     esi, [edx + sizeof.TCP_header]

  .opt_loop:

        cmp     esi, ecx                        ; are we scanning outside of header?

        jae     .no_options

        lodsb

        cmp     al, TCP_OPT_EOL                 ; end of option list?

        je      .no_options

        cmp     al, TCP_OPT_NOP

        je      .opt_loop

        cmp     al, TCP_OPT_MAXSEG

        je      .opt_maxseg

        cmp     al, TCP_OPT_WINDOW

        je      .opt_window

        cmp     al, TCP_OPT_SACK_PERMIT

        je      .opt_sack_permit

;        cmp     al, TCP_OPT_SACK

;        je      .opt_sack

        cmp     al, TCP_OPT_TIMESTAMP

        je      .opt_timestamp

        DEBUGF  DEBUG_NETWORK_VERBOSE, "TCP_input: unknown option:%u\n", al

        jmp     .no_options                     ; If we reach here, some unknown options were received, skip them all!

  .opt_maxseg:

        lodsb

        cmp     al, 4

        jne     .no_options                     ; error occured, ignore all options!

        test    [edx + TCP_header.Flags], TH_SYN

        jz      @f

        xor     eax, eax

        lodsw

        rol     ax, 8

        DEBUGF  DEBUG_NETWORK_VERBOSE, "TCP_input: Maxseg=%u\n", eax

        call    tcp_mss

       @@:

        jmp     .opt_loop

  .opt_window:

        lodsb

        cmp     al, 3

        jne     .no_options

        test    [edx + TCP_header.Flags], TH_SYN

        jz      @f

        DEBUGF  DEBUG_NETWORK_VERBOSE, "TCP_input: Got window scale option\n"

        or      [ebx + TCP_SOCKET.t_flags], TF_RCVD_SCALE

        lodsb

        mov     [ebx + TCP_SOCKET.SND_SCALE], al

        ;;;;; TODO

       @@:

        jmp     .opt_loop

  .opt_sack_permit:

        lodsb

        cmp     al, 2

        jne     .no_options

        test    [edx + TCP_header.Flags], TH_SYN

        jz      @f

        DEBUGF  DEBUG_NETWORK_VERBOSE, "TCP_input: Selective Acknowledgement permitted\n"

        or      [ebx + TCP_SOCKET.t_flags], TF_SACK_PERMIT

       @@:

        jmp     .opt_loop

  .opt_timestamp:

        lodsb

        cmp     al, 10                          ; length must be 10

        jne     .no_options

        DEBUGF  DEBUG_NETWORK_VERBOSE, "TCP_input: Got timestamp option\n"

        test    [edx + TCP_header.Flags], TH_SYN

        jz      @f

        or      [ebx + TCP_SOCKET.t_flags], TF_RCVD_TSTMP

       @@:

        lodsd

        bswap   eax

        mov     [ebx + TCP_SOCKET.ts_val], eax

        lodsd                                   ; timestamp echo reply

        mov     [ebx + TCP_SOCKET.ts_ecr], eax

        or      [temp_bits], TCP_BIT_TIMESTAMP

        ; Since we have a timestamp, lets do the paws test right away!

        test    [edx + TCP_header.Flags], TH_RST

        jnz     .no_paws

        mov     eax, [ebx + TCP_SOCKET.ts_recent]

        test    eax, eax

        jz      .no_paws

        cmp     eax, [ebx + TCP_SOCKET.ts_val]

        jbe     .no_paws

        DEBUGF  DEBUG_NETWORK_VERBOSE, "TCP_input: PAWS: detected an old segment\n"

        mov     eax, [timestamp]

        sub     eax, [ebx + TCP_SOCKET.ts_recent_age]

        pop     ecx

        cmp     eax, TCP_PAWS_IDLE

        jle     .paws_drop

        push    ecx

        mov     [ebx + TCP_SOCKET.ts_recent], 0         ; timestamp was invalid, fix it.

  .no_paws:

        jmp     .opt_loop

  .paws_drop:

        inc     [TCPS_rcvduppack]                       ; update stats

        add     [TCPS_rcvdupbyte], ecx

        inc     [TCPS_pawsdrop]

        jmp     .drop_after_ack

  .no_options:

        pop     ecx

;-----------------------------------------------------------------------

; 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_header.Flags], TH_SYN + TH_FIN + TH_RST + TH_URG

        jnz     .not_uni_xfer

        test    [edx + TCP_header.Flags], TH_ACK

        jz      .not_uni_xfer

        mov     eax, [edx + TCP_header.SequenceNumber]

        cmp     eax, [ebx + TCP_SOCKET.RCV_NXT]

        jne     .not_uni_xfer

        mov     eax, dword [edx + TCP_header.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]

        jb      .not_uni_xfer

; - The acknowledgment field in the segment is less than or equal to the maximum sequence number sent.

        mov     eax, [edx + TCP_header.AckNumber]

        cmp     eax, [ebx + TCP_SOCKET.SND_MAX]

        ja      .not_uni_xfer

; - The acknowledgment field in the segment is greater than the largest unacknowledged sequence number.

        sub     eax, [ebx + TCP_SOCKET.SND_UNA]

        jbe     .not_uni_xfer

        DEBUGF  DEBUG_NETWORK_VERBOSE, "TCP_input: Header prediction: we are sender\n"

;---------------------------------

; Packet is a pure ACK, process it

; Delete acknowledged bytes from send buffer

        pusha

        mov     ecx, eax

        lea     eax, [ebx + STREAM_SOCKET.snd]

        call    socket_ring_free

        popa

; Update RTT estimators

        test    [temp_bits], TCP_BIT_TIMESTAMP

        jz      .no_timestamp_rtt

        mov     eax, [timestamp]

        sub     eax, [ebx + TCP_SOCKET.ts_ecr]

        inc     eax

        call    tcp_xmit_timer

        jmp     .rtt_done

  .no_timestamp_rtt:

        cmp     [ebx + TCP_SOCKET.t_rtt], 0

        je      .rtt_done

        mov     eax, [edx + TCP_header.AckNumber]

        cmp     eax, [ebx + TCP_SOCKET.t_rtseq]

        jbe     .rtt_done

        mov     eax, [ebx + TCP_SOCKET.t_rtt]

        call    tcp_xmit_timer

  .rtt_done:

; update window pointers

        mov     eax, [edx + TCP_header.AckNumber]

        mov     [ebx + TCP_SOCKET.SND_UNA], eax

; Stop retransmit timer

        and     [ebx + TCP_SOCKET.timer_flags], not timer_flag_retransmission

; Unlock the socket

        pusha

        lea     ecx, [ebx + SOCKET.mutex]

        call    mutex_unlock

        popa

; Awaken waiting processes

        mov     eax, ebx

        call    socket_notify

; Generate more output

        call    tcp_output

        jmp     .drop_no_socket

;-------------------------------------------------

; 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_header.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.

        cmp     [ebx + TCP_SOCKET.seg_next], 0

        jne     .not_uni_xfer

; Complete processing of received data

        DEBUGF  DEBUG_NETWORK_VERBOSE, "TCP_input: Header prediction: we are receiving %u bytes\n", ecx

        mov     esi, [dataoffset]

        add     esi, edx

        lea     eax, [ebx + STREAM_SOCKET.rcv]

        call    socket_ring_write                       ; Add the data to the socket buffer

        add     [ebx + TCP_SOCKET.RCV_NXT], ecx         ; Update sequence number with number of bytes we have copied

        mov     eax, ebx

        call    socket_notify

        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  DEBUG_NETWORK_VERBOSE, "TCP_input: Header prediction failed\n"

; Calculate receive window size

        push    edx

        mov     eax, SOCKET_MAXDATA

        sub     eax, [ebx + STREAM_SOCKET.rcv.size]

        DEBUGF  DEBUG_NETWORK_VERBOSE, "Space in receive buffer=%d\n", eax

        mov     edx, [ebx + TCP_SOCKET.RCV_ADV]

        sub     edx, [ebx + TCP_SOCKET.RCV_NXT]

        DEBUGF  DEBUG_NETWORK_VERBOSE, "Current advertised window=%d\n", edx

        cmp     eax, edx

        jg      @f

        mov     eax, edx

       @@:

        DEBUGF  DEBUG_NETWORK_VERBOSE, "Receive window size=%d\n", eax

        mov     [ebx + TCP_SOCKET.RCV_WND], eax

        pop     edx

; If we are in listen or syn_sent state, go to that specific code right away

        cmp     [ebx + TCP_SOCKET.t_state], TCPS_LISTEN

        je      .LISTEN

        cmp     [ebx + TCP_SOCKET.t_state], TCPS_SYN_SENT

        je      .SYN_SENT

;----------------------------

; trim any data not in window

; 1. Check for duplicate data at beginning of segment

; Calculate number of bytes we need to drop

        mov     eax, [ebx + TCP_SOCKET.RCV_NXT]

        sub     eax, [edx + TCP_header.SequenceNumber]

        jle     .no_duplicate

        DEBUGF  DEBUG_NETWORK_VERBOSE, "TCP_input: %u bytes duplicate data!\n", eax

        test    [edx + TCP_header.Flags], TH_SYN

        jz      .no_dup_syn

        DEBUGF  DEBUG_NETWORK_VERBOSE, "TCP_input: got duplicate syn\n"

        and     [edx + TCP_header.Flags], not (TH_SYN)

        inc     [edx + TCP_header.SequenceNumber]

        cmp     [edx + TCP_header.UrgentPointer], 1

        jbe     @f

        dec     [edx + TCP_header.UrgentPointer]

        jmp     .dup_syn

       @@:

        and     [edx + TCP_header.Flags], not (TH_URG)

  .dup_syn:

        dec     eax

  .no_dup_syn:

; 2. Check for entire duplicate segment

        cmp     eax, ecx                ; eax holds number of bytes to drop, ecx is data size

        jb      .duplicate

        jnz     @f

        test    [edx + TCP_header.Flags], TH_FIN

        jnz     .duplicate

       @@:

; Any valid FIN must be to the left of the window.

; At this point the FIN must be out of sequence or a duplicate, drop it

        and     [edx + TCP_header.Flags], not TH_FIN

; send an ACK and resynchronize and drop any data.

; But keep on processing for RST or ACK

        or      [ebx + TCP_SOCKET.t_flags], TF_ACKNOW

        mov     eax, ecx

        inc     [TCPS_rcvpartduppack]

;;; TODO: update stats

;-----------------------------------------------

; Remove duplicate data and update urgent offset

  .duplicate:

        DEBUGF  DEBUG_NETWORK_VERBOSE, "TCP_input: trimming duplicate data\n"

; Trim data from left side of window

        add     [dataoffset], eax

        add     [edx + TCP_header.SequenceNumber], eax

        sub     ecx, eax

        sub     [edx + TCP_header.UrgentPointer], ax

        jg      @f

        and     [edx + TCP_header.Flags], not (TH_URG)

        mov     [edx + TCP_header.UrgentPointer], 0

       @@:

;--------------------------------------------------

; Handle data that arrives after process terminates

  .no_duplicate:

        cmp     [ebx + SOCKET.PID], 0                   ;;; TODO: use socket flags instead??

        jne     .not_terminated

        cmp     [ebx + TCP_SOCKET.t_state], TCPS_CLOSE_WAIT

        jbe     .not_terminated

        test    ecx, ecx

        jz      .not_terminated

        mov     eax, ebx

        call    tcp_close

        inc     [TCPS_rcvafterclose]

        jmp     .respond_seg_reset

;----------------------------------------

; Remove data beyond right edge of window

  .not_terminated:

        mov     eax, [edx + TCP_header.SequenceNumber]

        add     eax, ecx

        sub     eax, [ebx + TCP_SOCKET.RCV_NXT]

        sub     eax, [ebx + TCP_SOCKET.RCV_WND]         ; eax now holds the number of bytes to drop

        jle     .no_excess_data

        DEBUGF  DEBUG_NETWORK_VERBOSE, "%d bytes beyond right edge of window\n", eax

;;; TODO: update stats

        cmp     eax, ecx

        jl      .dont_drop_all

; If a new connection request is received while in TIME_WAIT, drop the old connection and start over,

; if the sequence numbers are above the previous ones

        test    [edx + TCP_header.Flags], TH_SYN

        jz      .no_new_request

        cmp     [ebx + TCP_SOCKET.t_state], TCPS_TIMED_WAIT

        jne     .no_new_request

;        mov     edx, [ebx + TCP_SOCKET.RCV_NXT]

;        cmp     edx, [edx + TCP_header.SequenceNumber]

;        add     edx, 64000      ; TCP_ISSINCR   FIXME

        mov     eax, ebx

        call    tcp_close

        jmp     .findpcb        ; FIXME: skip code for unscaling window, ...

  .no_new_request:

; If window is closed, we can only take segments at window edge, and have to drop data and PUSH from

; incoming segments. Continue processing, but remember to ACK. Otherwise drop segment and ACK

        cmp     [ebx + TCP_SOCKET.RCV_WND], 0

        jne     .drop_after_ack

        mov     esi, [edx + TCP_header.SequenceNumber]

        cmp     esi, [ebx + TCP_SOCKET.RCV_NXT]

        jne     .drop_after_ack

        or      [ebx + TCP_SOCKET.t_flags], TF_ACKNOW

;;; TODO: update stats

  .dont_drop_all:

;;; TODO: update stats

        DEBUGF  DEBUG_NETWORK_VERBOSE, "Trimming %u bytes from the right of the window\n"

        sub     ecx, eax        ; remove data from the right side of window (decrease data length)

        and     [edx + TCP_header.Flags], not (TH_PUSH or TH_FIN)

  .no_excess_data:

;-----------------

; Record timestamp

; If last ACK falls within this segments sequence numbers, record its timestamp

        test    [temp_bits], TCP_BIT_TIMESTAMP

        jz      .no_timestamp

        mov     eax, [ebx + TCP_SOCKET.last_ack_sent]

        sub     eax, [edx + TCP_header.SequenceNumber]

        jb      .no_timestamp

        test    [ebx + TCP_header.Flags], TH_SYN or TH_FIN      ; syn and fin occupy one byte

        jz      @f

        dec     eax

       @@:

        sub     eax, ecx

        jae     .no_timestamp

        DEBUGF  DEBUG_NETWORK_VERBOSE, "Recording timestamp\n"

        mov     eax, [timestamp]

        mov     [ebx + TCP_SOCKET.ts_recent_age], eax

        mov     eax, [ebx + TCP_SOCKET.ts_val]

        mov     [ebx + TCP_SOCKET.ts_recent], eax

  .no_timestamp:

;------------------

; Process RST flags

        test    [edx + TCP_header.Flags], TH_RST

        jz      .no_rst

        DEBUGF  DEBUG_NETWORK_VERBOSE, "TCP_input: Got an RST flag\n"

        mov     eax, [ebx + TCP_SOCKET.t_state]

        shl     eax, 2

        jmp     dword [eax + .rst_sw_list]

  .rst_sw_list:

        dd      .no_rst         ; TCPS_CLOSED

        dd      .no_rst         ; TCPS_LISTEN

        dd      .no_rst         ; 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  DEBUG_NETWORK_VERBOSE, "TCP_input: Connection refused\n"

        mov     [ebx + SOCKET.errorcode], ECONNREFUSED

        jmp     .close

  .econnreset:

        DEBUGF  DEBUG_NETWORK_VERBOSE, "TCP_input: Connection reset\n"

        mov     [ebx + SOCKET.errorcode], ECONNRESET

  .close:

        DEBUGF  DEBUG_NETWORK_VERBOSE, "TCP_input: Closing connection\n"

        mov     [ebx + TCP_SOCKET.t_state], TCPS_CLOSED

;;; TODO: update stats (tcp drops)

        mov     eax, ebx

        call    tcp_close

        jmp     .drop_no_socket

  .rst_close:

        DEBUGF  DEBUG_NETWORK_VERBOSE, "TCP_input: Closing with reset\n"

        mov     eax, ebx

        call    tcp_close

        jmp     .drop_no_socket

  .no_rst:

;--------------------------------------

; handle SYN-full and ACK-less segments

        test    [edx + TCP_header.Flags], TH_SYN

        jz      .not_syn_full

        mov     eax, ebx

        mov     ebx, ECONNRESET

        call    tcp_drop

        jmp     .drop_with_reset

  .not_syn_full:

;---------------

; ACK processing

        test    [edx + TCP_header.Flags], TH_ACK

        jz      .drop

        cmp     [ebx + TCP_SOCKET.t_state], TCPS_SYN_RECEIVED

        jb      .ack_processed                                  ; states: closed, listen, syn_sent

        ja      .no_syn_rcv                                     ; established, fin_wait_1, fin_wait_2, close_wait, closing, last_ack, time_wait

        DEBUGF  DEBUG_NETWORK_VERBOSE, "TCP_input: state=syn_received\n"

        mov     eax, [edx + TCP_header.AckNumber]

        cmp     [ebx + TCP_SOCKET.SND_UNA], eax

        ja      .drop_with_reset

        cmp     eax, [ebx + TCP_SOCKET.SND_MAX]

        ja      .drop_with_reset

;;; TODO: 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

        push    word [ebx + TCP_SOCKET.requested_s_scale]       ; Set send and receive scale factors to the received values

        pop     word [ebx + TCP_SOCKET.SND_SCALE]

       @@:

        call    tcp_reassemble

        mov     eax, [edx + TCP_header.SequenceNumber]

        dec     eax

        mov     [ebx + TCP_SOCKET.SND_WL1], eax

  .no_syn_rcv:

;-------------------------

; check for duplicate ACKs

        mov     eax, [edx + TCP_header.AckNumber]

        cmp     eax, [ebx + TCP_SOCKET.SND_UNA]

        ja      .not_dup_ack

        test    ecx, ecx

        jnz     .reset_dupacks

        mov     eax, dword [edx + TCP_header.Window]

        cmp     eax, [ebx + TCP_SOCKET.SND_WND]

        jne     .reset_dupacks

        DEBUGF  DEBUG_NETWORK_VERBOSE, "TCP_input: Processing duplicate ACK\n"

; If we have outstanding data, other than a window probe, this is a completely duplicate ACK

; (window info didnt change) The ACK is the biggest we've seen and we've seen exactly our rexmt threshold of them,

; assume a packet has been dropped and retransmit it. Kludge snd_nxt & the congestion window so we send only this one packet.

        test    [ebx + TCP_SOCKET.timer_flags], timer_flag_retransmission

        jz      @f

        mov     eax, [edx + TCP_header.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]

        jbe     @f

        mov     eax, [ebx + TCP_SOCKET.SND_CWND]

  @@:

        shr     eax, 1

        push    edx

        xor     edx, edx

        div     [ebx + TCP_SOCKET.t_maxseg]

        cmp     eax, 2

        ja      @f

        xor     eax, eax

        mov     al, 2

       @@:

        mul     [ebx + TCP_SOCKET.t_maxseg]

        pop     edx

        mov     [ebx + TCP_SOCKET.SND_SSTHRESH], eax

        and     [ebx + TCP_SOCKET.timer_flags], not timer_flag_retransmission   ; turn off retransmission timer

        mov     [ebx + TCP_SOCKET.t_rtt], 0

        mov     eax, [edx + TCP_header.AckNumber]

        mov     [ebx + TCP_SOCKET.SND_NXT], eax

        mov     eax, [ebx + TCP_SOCKET.t_maxseg]

        mov     [ebx + TCP_SOCKET.SND_CWND], eax

; Unlock the socket

        push    ebx

        lea     ecx, [ebx + SOCKET.mutex]

        call    mutex_unlock

; retransmit missing segment

        mov     eax, [esp]

        call    tcp_output

; Lock the socket again

        mov     ecx, [esp]

        add     ecx, SOCKET.mutex

        call    mutex_lock

        pop     ebx

; Continue processing

        xor     edx, edx

        mov     eax, [ebx + TCP_SOCKET.t_maxseg]

        mul     [ebx + TCP_SOCKET.t_dupacks]

        add     eax, [ebx + TCP_SOCKET.SND_SSTHRESH]

        mov     [ebx + TCP_SOCKET.SND_CWND], eax

        pop     eax                                     ; <<<<

        cmp     eax, [ebx + TCP_SOCKET.SND_NXT]

        jb      @f

        mov     [ebx + TCP_SOCKET.SND_NXT], eax

       @@:

        jmp     .drop

  .no_re_xmit:

        jbe     .not_dup_ack

        DEBUGF  DEBUG_NETWORK_VERBOSE, "TCP_input: Increasing congestion window\n"

        mov     eax, [ebx + TCP_SOCKET.t_maxseg]

        add     [ebx + TCP_SOCKET.SND_CWND], eax

; Unlock the socket

        push    ebx

        lea     ecx, [ebx + SOCKET.mutex]

        call    mutex_unlock

; retransmit missing segment

        mov     eax, [esp]

        call    tcp_output

; Lock the socket again

        mov     ecx, [esp]

        add     ecx, SOCKET.mutex

        call    mutex_lock

        pop     ebx

        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]

        ja      @f

        cmp     [ebx + TCP_SOCKET.t_dupacks], TCP_re_xmit_thresh

        jbe     @f

        mov     [ebx + TCP_SOCKET.SND_CWND], eax

       @@:

        mov     [ebx + TCP_SOCKET.t_dupacks], 0

        mov     eax, [edx + TCP_header.AckNumber]

        cmp     eax, [ebx + TCP_SOCKET.SND_MAX]

        jbe     @f

        inc     [TCPS_rcvacktoomuch]

        jmp     .drop_after_ack

       @@:

        mov     edi, [edx + TCP_header.AckNumber]

        sub     edi, [ebx + TCP_SOCKET.SND_UNA]         ; now we got the number of acked bytes in edi

        inc     [TCPS_rcvackpack]

        add     [TCPS_rcvackbyte], edi

        DEBUGF  DEBUG_NETWORK_VERBOSE, "TCP_input: acceptable ACK for %u bytes\n", edi

;------------------------------------------

; RTT measurements and retransmission timer

; If we have a timestamp, update smoothed RTT

        test    [temp_bits], TCP_BIT_TIMESTAMP

        jz      .timestamp_not_present

        mov     eax, [timestamp]

        sub     eax, [ebx + TCP_SOCKET.ts_ecr]

        inc     eax

        call    tcp_xmit_timer

        jmp     .rtt_done_

; If no timestamp but transmit timer is running and timed sequence number was acked,

; update smoothed RTT. Since we now have an RTT measurement, cancel the timer backoff

; (Phil Karn's retransmit algo)

; Recompute the initial retransmit timer

  .timestamp_not_present:

        mov     eax, [edx + TCP_header.AckNumber]

        cmp     eax, [ebx + TCP_SOCKET.t_rtseq]

        jbe     .rtt_done_

        mov     eax, [ebx + TCP_SOCKET.t_rtt]

        test    eax, eax

        jz      .rtt_done_

        call    tcp_xmit_timer

  .rtt_done_:

; If all outstanding data is acked, stop retransmit timer and remember to restart (more output or persist)

; If there is more data to be acked, restart retransmit timer, using current (possible backed-off) value.

        mov     eax, [ebx + TCP_SOCKET.SND_MAX]

        cmp     eax, [edx + TCP_header.AckNumber]

        jne     .more_data

        and     [ebx + TCP_SOCKET.timer_flags], not timer_flag_retransmission

        or      [temp_bits], TCP_BIT_NEEDOUTPUT

        jmp     .no_restart

  .more_data:

        test    [ebx + TCP_SOCKET.timer_flags], timer_flag_persist

        jnz     .no_restart

        mov     eax, [ebx + TCP_SOCKET.t_rxtcur]

        mov     [ebx + TCP_SOCKET.timer_retransmission], eax

        or      [ebx + TCP_SOCKET.timer_flags], timer_flag_retransmission

  .no_restart:

;-------------------------------------------

; 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]

        jbe     @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

        jbe     @f

        mov     esi, eax

  @@:

        mov     [ebx + TCP_SOCKET.SND_CWND], esi

;------------------------------------------

; Remove acknowledged data from send buffer

        cmp     edi, [ebx + STREAM_SOCKET.snd.size]

        jbe     .finiacked

        push    ecx edx ebx

        mov     ecx, [ebx + STREAM_SOCKET.snd.size]

        lea     eax, [ebx + STREAM_SOCKET.snd]

        sub     [ebx + TCP_SOCKET.SND_WND], ecx

        call    socket_ring_free

        pop     ebx edx ecx

        DEBUGF  DEBUG_NETWORK_VERBOSE, "TCP_input: our FIN is acked\n"

        or      [temp_bits], TCP_BIT_FIN_IS_ACKED