[tipc-discussion] Update on Solaris port for TIPC

[Stephens, A. <all...@wi...>]

Reposted to general TIPC mailing list for wider circulation.

-----Original Message-----
From: Lau...@su... [mailto:Lau...@su...]=20
Sent: Tuesday, December 18, 2007 1:58 PM
To: Stephens, Allan; Horvath, Elmer
Cc: Renaud M=E9trich
Subject: Re: [tipc-discussion] REMINDER - TIPC Working Group Meeting on =
Thursday


> PROPOSED AGENDA:
>
> - TIPC on Solaris progress report

Hi Al and Elmer,

I'm in vacation in a few minutes, but Renaud M=E9trich will be able to =
attend the meeting.
Renaud investigated how to implement the TIPC socket module on Solaris =
10, and knows what I did while porting on OpenSolaris.

My objective was to have most of the code compiling on solaris, and try =
to get 'tipc-config -s' return something, before I went to vacation. And =
it returns something !

$ tipc-config -s
invalid reply message received via TIPC

Not what I expected :-)
But part of it went well :
The tipc-config command opens a socket, calls setsockopt() and sendto() =
to send a message TIPC_CMD_SHOW_STATS to the socket.
The message is received by our socket kernel module,
tipc_port_recv_msg() is called, calls port_dispatcher() that schedules a =
task.

Then on the tipc-config side, sendto() returns, and tipc-config calls =
poll().

The kernel is executing the task that was preciously scheduled by =
tipc_port_recv_msg().
port_dispatcher_sigh() sees that it is a TIPC_NAMED_MSG and calls the =
callback cfg_named_msg_event() which calls tipc_show_stats() and send =
the reply by calling tipc_send_buf2port().
Finally the callback dispatch() from our socket module is called, it =
will enqueue the message and do a poll wakeup.

tipc-config is awaken and calls recv(). The kernel module dequeues the =
message, transfers it to a specific user io structure.
tipc-config receives it in an answer , but header contents like len and =
type have wrong values.

I still need to debug this.
The vxworks code was very helpfull for the sk_buff management. But there =
are some more pointers in mblk_t solaris structures that I surely didn't =
manage correctly somewhere. Or the problem may be in what I did to =
transfer the kernel data to the specific uio structure.

Apart from that, I've coded the tipc_eth_media but not tested.
And several socket callbacks aren't yet coded : accept, bind, listen, =
connect...
Still some work for January !

Have a nice break
Bye