Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!uupsi!sunic!isgate!krafla!einari
From: einari@rhi.hi.is (Einar Indridason)
Newsgroups: comp.os.minix
Subject: Re: (PC/AT) why not include 'sync' in keyboard.c (for rebooting?)
Message-ID: <3058@krafla.rhi.hi.is>
Date: 17 Apr 91 17:17:27 GMT
References: <50809@nigel.ee.udel.edu> <1991Apr15.210247.26781@cmcl2.nyu.edu>
Reply-To: einari@rhi.hi.is (Einar Indridason)
Organization: University of Iceland (RHI)
Lines: 52

In article <1991Apr15.210247.26781@cmcl2.nyu.edu> jai@lab.ultra.nyu.edu (Benchiao Jai) writes:
>The kernel can send a message to fs telling it to execute a sync, but
>this operation is not atomic, so other things may happen in between.
>Picture this:	keyboard driver send sync request to fs
>		fs sends a write request to disk task
>		disk task executes the request and send SUSPEND to fs
>		fs suspends keyboard task and fetches another request from user
>		(this is possible since no higher priority tasks can be run
>		 at this moment)
>		fs executes the new write request by putting the new data
>		  into buffer cache
>		disk task replies OK
>		fs revives keyboard task and send an OK reply
>		keyboard proceeds to reboot
>BOOM!  Buffer cache is still dirty.
>



Ok.  I understand.  Since it was me who started this discussion I might
say something :-)

Well, to start with:  fifteen minutes after I posted my query I thought
of a situation that could be invalid.  It is in the same lines as the
above quotation is.

(We had a disk crash shortly after I posted my query, so I might have
missed something of the early discussion of it :-/


In short: My thought was not quite correct.  It is *not* ok to put
'sync()' in 'keyboard.c' to flush buffers to the filesystem before
rebooting.

Thanks to those who pointed it to me.  One of them suggested the
following solution:  (usable if you are the only user on the machine and
not neccessary using the virtual consoles patch, (which I haven't
installed yet))
	Create an account with a login shell composed of a simple C
program that contain a loop surrounding a 'sync()' call.
	Wait until all drive-access lights have gone out.
	Boot the machine.


Well, thats enough for now.  (You just wait for my next question :-)


--
Internet:    einari@rhi.hi.is        |   "Just give me my command line and drag
UUCP:    ..!mcsun!isgate!rhi!einari  |   the GUIs to the waste basket!!!!"

Surgeon Generals warning:  Masking the 8th bit can seriously damage your brain!!