From mboxrd@z Thu Jan  1 00:00:00 1970
Message-ID: <6c7727bd06577062ffdbadcef28e33bf@rei2.9hal>
Date: Fri, 15 Mar 2013 21:35:49 +0100
From: cinap_lenrek@gmx.de
To: 9fans@9fans.net
In-Reply-To: <aab086c1caa3c5fbe10219ef85ca601f@ladd.quanstro.net>
MIME-Version: 1.0
Content-Type: text/plain; charset="US-ASCII"
Content-Transfer-Encoding: 7bit
Subject: Re: [9fans] exec() question
Topicbox-Message-UUID: 2af739f0-ead8-11e9-9d60-3106f5b1d025

i think you are right. the temporary tstk segment will be before
the stack segment like:

| txt | dat | bss ... | *unmapedspace* | tstk | stk |

the segattach syscall only makes sure you dont map something after
or overlapping with the stack. so i think you could indeed map
something there and make the front fall off after exec().

you could map a readonly segment there and make the kernel crash
when it tries prepare the new stack.

segattach() also would try to allocate below the lowest possible stack
address when you pass 0 as the address.

maybe the tstk (ESEG) should be placed *after* the stack swaping
tstk and stk like:

#define	TSTKTOP		(VMAP-BY2PG)
#define	TSTKSIZ		100
#define	USTKTOP		(TSTKTOP-TSTKSIZ)
#define	USTKSIZE	(16*1024*1024)

but maybe just making the checks in segattach take the tstk into
account is simpler...

--
cinap