paralell multipart sms not working

[Dennis Wehrle]

Hi

>
>> until here everything is OK, except that the send-function should
return the
>> number of bytes which are sent, but the value is always too large
for three.
>
> so it returns less than what we wrote?
 >

No msg->len and ret are equal. I think it was my fault. I had count the 
bytes manually. For example:

<001b> ipaccess.c:512 TX 2: 03 01 01 28 02 03 0b 00 02 49 04
<001b> ipaccess.c:514 msg->len 14
<001b> ipaccess.c:520 bfd->fd: 12
<001b> ipaccess.c:521 ret: 14

On line 512 i count manually 15 bytes, but the msg->len and ret value is 18.
I think this is OK, than the bytes on line 512 doesn't contain the IPA 
protocol bytes. I could verify this on wireshark. There are 3 bytes more 
for the IPA header (00 0b 00). So i would say that i've misinterpreted 
that. On line 510 the function ipaccess_prepend_header is called and 
therefore i thought that the bytes for the header is included at the 
output. I didn't concerned that only the msg->l2h is printed.

 >
> You can set a delay timer to not things as fast as possible and
probably get IPA packets into different IP packets.
>

Unfortunately this doesn't solve the problem.

 >
> this is valid behavior (and you see it during RSL/OML bring up), one
thing how
 >

Ok. If this is a valid behavior, maybe i'm on the wrong track. Therefore 
i don't know how to investigate the problem further.

 >
> this can break (and would break OpenBSC/libosmo-abis) is that we do:
>
> 	1.) ipa_header = recv(3 bytes)
> 	2.) data = recv(ipa_header->len)
>
> this would fail in case header and data do not appear on the same
frame. So
> the esoteric question is do you have MTU changes in your network
(this assumes
> that the ipaccess would have similar broken code we have)?
>

No the standard MTU (1500 Bytes) is set. Even if i connect it directly 
on my pc via a switch the problem occurs. My abis pcap file was created 
this way.


Best regards
Dennis Wehrle