Skip site navigation (1)Skip section navigation (2)
Date:      Mon, 27 Jan 2003 00:35:36 -0800
From:      Marcel Moolenaar <marcel@xcllnt.net>
To:        phk@freebsd.org
Cc:        Nate Lawson <nate@root.org>, cvs-committers@freebsd.org, cvs-all@freebsd.org
Subject:   Re: cvs commit: src/sbin/disklabel disklabel.c
Message-ID:  <20030127083536.GA2644@dhcp01.pn.xcllnt.net>
In-Reply-To: <19313.1043652737@critter.freebsd.dk>
References:  <20030127070958.GA2431@dhcp01.pn.xcllnt.net> <19313.1043652737@critter.freebsd.dk>

next in thread | previous in thread | raw e-mail | index | archive | help
On Mon, Jan 27, 2003 at 08:32:17AM +0100, phk@freebsd.org wrote:
> In message <20030127070958.GA2431@dhcp01.pn.xcllnt.net>, Marcel Moolenaar write
> s:
> 
> >One of the thoughts I have on the issue is to have the geom classes
> >have device nodes as side-entrances, independent of the producer-
> >consumer interfaces. Direct access to the device nodes can then
> >easily be checked with each read or write without pessimizing the
> >P/C layering.
> >
> >The ioctl interface not not suited for GPT as the amount of data that
> >has to be read or written is not fixed, as are the LBAs at which the
> >data has to be written.
> 
> Well, that doesn't really mean that ioctl cannot be used, merely
> that the arguments have to be taylored.

True, but why do we want to create an artificial interface on top
of ioctl(2) or by means of ioctl(2), that's less suitable than the
existing read(2), write(2) and lseek(2) mechanism, which BTW is in
use by gpt(8) already?

When we talked about this more than 6 months ago (IIRC) you suggested
the use of a sysctl (which is roughly equivalent to an ioctl) to
enable foot-shooting. That would be an acceptable solution IMO. Using
ioctl (or sysctl for that matter) to implemement an alternate read-
write interface for single sector boot blocks at fixed LBAs is getting
ugly, but one could get away with it. Trying to implement an alternate
read/write interface that's as flexible as existing interfaces to
handle GPT is a bad idea and volates POLA. And all this because we
haven't got the infrastructure in the kernel to deal with permissions
without pessimizing disk I/O? Sorry, that doesn't seem right to me...

-- 
 Marcel Moolenaar	  USPA: A-39004		 marcel@xcllnt.net

To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe cvs-all" in the body of the message




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20030127083536.GA2644>