(usagi-users 03386) Re: MLDv2 linux implementation questions

Yes, download :

http://clarinet.u-strasbg.fr/~hoerdt/test.tgz

uncompress it, and compile it (make) and launch :

test_msf ff3e::1234 < gen.t

gen.t contains instruction for test_msf, alternating joining and leaving
the same
SSM channel using the delta based API (struct group_source_req). At the
beginning
you will see that joining is OK, but after some alternate, the kernel
returns that
no more memory is available. I tested it with net.core.optmem_max =
10240 on a
linux 2.6.11

Hoerdt Mickaël


David Stevens wrote:

>Hoerdt Mickael <[email protected]> wrote on 06/13/2005 11:32:37 
>AM:
>
>  
>
>>If I count right, the 4th field is im->mca_users, which is not the time
>>left on the pending
>>timer, but I understand right, the number of setsockopt() for this
>>group. The thing
>>that I don't understand is that when I join a channel (Include mode
>>using ADD_SOURCE) and leave it alternavely, this value increase when it
>>should stay constant or is there an optimisation ?
>>    
>>
>
>
>mca_users is the number of references the kernel has to that multicast
>group on the device. New joins should increment, leaves should
>decrement it, the same as in MLDv1.
>
>The multicast version in use is internal, and not exported currently.
>It seems like a reasonable addition to igmp6 to me (igmp for ipv4 has
>similar).
>
>In another mail, you mentioned a sock buffer issue; if you can send
>me your program (which I hope is small :-) ), I can take a look.
>
>                                                        +-DLS
>
>
>  
>