[Click] Problem with KernelTap : size of packet readed from TAP

Eddie Kohler kohler at cs.ucla.edu
Tue Jul 22 18:01:18 EDT 2008


Hi Remi,

Thanks very much for writing!  I think I've fixed the KernelTap bug you 
reported, using this patch:

http://read.cs.ucla.edu/gitweb?p=click;a=blobdiff;f=elements/userlevel/kerneltun.cc;h=e897364a31bed3e11bf512babf3cd84548d27e58;hp=293068c3664760612e7dbb43e34a2b88565e9ce6;hb=40151bcca1feae2ed2ff471fd55b828bc988fad8;hpb=dee5bcf884b7670b8a3926c37b2d1966b1b092eb

I hear you about the documentation of cp_va_parse(), and went ahead and wrote 
a transition guide for how to update cp_va_parse() to cp_va_kparse().  It is here:

http://www.read.cs.ucla.edu/click/doxygen/confparse_8hh.html#40f5799c9c2647e71460572c81b4a552

As to your other bugs, I'd like to fix them too.

* Do you have a test config showing how EtherSpanTree is broken?  What is the 
error?
* Do you have a test config showing how handler writes are broken?

Thanks!
Eddie


remi.clavier at orange-ftgroup.com wrote:
>  
> Many thanks, Eddy, to take my problem into account.
> I use the new version proposed and I experience that the 14 bytes of the Ethernet header are not read. But if I change the nummer of bytes to read (increasing it from _mtu_in to _mtu_in +14, increase the size of pcket and the dedicated "pack" command, all work fine with the new version and I'm now able to correctly make session (ssh, http,...) between two PC using click. Thank's a lot.
> When I try to use the full version of click in the repository I experience some other problems :
> 1)EtherSpanTree don't work in this version
> 2)Write something in an write switch handler seems to be work not excatly as in the used version ( write toto.switch 9 give sometimes an "integer value expected error (or something like that))
> Perhaps some other problem, I can' testt my click bridge more deeper
> 3) depreciated for ca_va_parse not documented for changes
> 
> So I use my old version of click (1.6.0) and only change the needed files (packet.hh, packet.cc, kerneltun.hh,kerneltun.cc) and it'seems to work fine for my (complex) ethernet bridges.
> 
> Another Question. Does an MSTP version of spanning tree exists? If yes, it's possible to have a pointer to a people able to help me to install it in click
> 
> Thanks a lot for all Your work for click users
> 
> Remi
> 
> -----Message d'origine-----
> De : Eddie Kohler [mailto:kohler at cs.ucla.edu] 
> Envoyé : jeudi 17 juillet 2008 00:49
> À : CLAVIER Remi RD-CORE-LAN
> Cc : click at amsterdam.lcs.mit.edu
> Objet : Re: [Click] Problem with KernelTap : size of packet readed from TAP
> 
> Remi,
> 
> It certainly would seem more natural to treat the MTU consistently as counting the network header only.  (It used to count network + link headers in the KernelTap case.)  I have checked in a patch which also fixes some other bugs. 
>   If you aren't using git, the diff can be downloaded here:
> 
> http://www.read.cs.ucla.edu/gitweb?p=click;a=commitdiff;h=6014975667600091c1992fca1c859c28bddfa869
> 
> Thanks for the bug report.
> Eddie
> 
> 
> remi.clavier at orange-ftgroup.com wrote:
>>  
>> Thanks to Michael Voorhaen [michael.voorhaen at ua.ac.be], the link 
>> between two TCP applications using click works. But, in my case, only 
>> for frames with a lesser size than MTU.
>> If frame size is equal at MTU (case of fragmented frame), 14 bytes are 
>> missing at the end of it.
>> A deeper analyze of the code thows that the number of read bytes is 
>> equal to _mtu_in (=MTU+2) and the variable _type is 0.
>> The size of the information to read is MTU+16... (
>> MTU+2bytes+EthernetHeader) At is easy to show if we increase the 3th
>> param value of the read (_fd,...) function.
>>
>> Perhaps it's a problem of compilation constant, or a bad behaviour of 
>> tap on my LINUX implementation (ubuntu 8.04).
>>
>> It's this a know issue and, if yes, it's a patch or a workaround about?
>>
>> Thanks for all help
>>
>> Regards
>>
>> Remi
>>
>> _______________________________________________
>> click mailing list
>> click at amsterdam.lcs.mit.edu
>> https://amsterdam.lcs.mit.edu/mailman/listinfo/click



More information about the click mailing list