kernel/network: fix ARP remove on devices other than 0

arp_api .remove forgot the per-device table offset, so it deleted
from device 0's table and shifted device 1's entries into it.

kernel/trunk/network/ARP.inc

@@ -688,10 +688,12 @@ arp_api:
         ; ecx = # entry
         cmp     ecx, [ARP_entries + eax]
         jae     .error
-        imul    ecx, sizeof.ARP_entry
-        lea     esi, [ARP_table + ecx]
         mov     edi, eax
         shr     edi, 2
+        mov     esi, edi
+        imul    esi, sizeof.ARP_entry*ARP_TABLE_SIZE
+        imul    ecx, sizeof.ARP_entry
+        lea     esi, [ARP_table + esi + ecx]
         call    arp_del_entry
         ret
 

kernel/trunk/network/tcp.inc

@@ -72,7 +72,7 @@ TCP_time_rtt_default    = 5           ; default Round Trip Time (3,2s)
 TCP_time_srtt_default   = 0           ;
 TCP_time_max_idle       = 8*TCP_time_keep_interval      ; FIXME
 
-TCP_time_connect        = 1000        ; in 1/100s (default=10s)
+TCP_time_connect        = 300         ; in 1/100s (default=3s)
 
 ; timer constants
 TCP_max_rxtshift        = 12          ; max retransmissions waiting for ACK

kernel/trunk/network/tcp_input.inc

@@ -399,7 +399,6 @@ endl
         bswap   eax
         mov     [ebx + TCP_SOCKET.ts_val], eax
         lodsd                                   ; timestamp echo reply
-        bswap   eax
         mov     [ebx + TCP_SOCKET.ts_ecr], eax
         or      [temp_bits], TCP_BIT_TIMESTAMP
 
@@ -979,7 +978,7 @@ endl
 
         inc     [ebx + TCP_SOCKET.t_dupacks]
         cmp     [ebx + TCP_SOCKET.t_dupacks], TCP_re_xmit_thresh
-        jb      .ack_processed
+        jb      .dup_ack_complete
         ja      .another_lost
 
         DEBUGF  DEBUG_NETWORK_VERBOSE, "TCP_input: Re-transmitting lost segment\n"

kernel/trunk/network/tcp_output.inc

@@ -391,14 +391,8 @@ endl
         jz      .no_timestamp
 
   .timestamp:
-        push    ebx
-        mov     ebx, [eax + TCP_SOCKET.ts_recent]
-        bswap   ebx
-        xchg    ebx, [esp]                      ; TSecr
-        push    ebx
-        mov     ebx, [timer_ticks]
-        bswap   ebx
-        xchg    ebx, [esp]                      ; TSval
+        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
 
@@ -419,12 +413,10 @@ endl
 ; check if we don't exceed the max segment size
 
         add     esi, edi                        ; total TCP segment size
-        mov     ebx, [eax + TCP_SOCKET.t_maxseg]
-        add     ebx, sizeof.TCP_header          ; MSS does not count the header
-        cmp     esi, ebx
+        cmp     esi, [eax + TCP_SOCKET.t_maxseg]
         jbe     .no_overflow
 
-        mov     esi, ebx
+        mov     esi, [eax + TCP_SOCKET.t_maxseg]
         or      [temp_bits], TCP_BIT_SENDALOT
   .no_overflow: