[prev in list] [next in list] [prev in thread] [next in thread] 

List:       osdl-storage-sig
Subject:    Re: [Storage_sig] Re: New release of Linux POSIX AIO
From:       Laurent Vivier <Laurent.Vivier () bull ! net>
Date:       2005-03-31 17:16:10
Message-ID: 1112289370.17272.6.camel () imac
[Download RAW message or body]

Le jeudi 31 mars 2005 à 11:06 -0500, Benjamin LaHaise a écrit :
> On Thu, Mar 31, 2005 at 05:23:59PM +0200, Laurent Vivier wrote:
> > > aio_reqprio and aio_lio_opcode.  Given the nature of priorities and 
> > > the number of opcodes, a short provides a more than ample range.
> > 
> > On this point, I totally agree with you... but it seems it's not point
> > of view of POSIX people.
> 
> In that case, why not have the fields that can overlap do and fixup the 
> remainder?  The resulting aiocb would look like:
> 
> 	struct aiocb {
> 		... first 64 bytes match iocb ...
> 		int aio_reqprio;
> 		int aio_lio_opcode;
> 		struct sigevent aio_sigevent;
> 	};
> 

I think it's a very good solution.

> > Put I think this needs two user-to-kernel copies, one for iocb and one
> > for sigevent.
> > It's why our libposix-aio doesn't pass sigevent to kernel but only three
> > useful values inside iocb when we create iocb from aiocb.
> 
> Sounds like a perfect job for __get_user_nocheck -- only copy the data 
> that is needed into the in-kernel structures after validating the pointer.

Thank you for the tip ;-)

Sébastien, do you agree with all of that ?

> 		-ben

Regards,
Laurent
-- 
Laurent Vivier <Laurent.Vivier@bull.net>




_______________________________________________
Storage_sig mailing list
Storage_sig@lists.osdl.org
http://lists.osdl.org/mailman/listinfo/storage_sig


[prev in list] [next in list] [prev in thread] [next in thread]