[35072] in bugtraq
Re: Linux Kernel sctp_setsockopt() Integer Overflow
daemon@ATHENA.MIT.EDU (Jirka Kosina)
Mon May 31 14:43:42 2004
Date: Sat, 29 May 2004 05:13:17 +0200 (CEST)
From: Jirka Kosina <jikos@jikos.cz>
To: Michael Tokarev <mjt@tls.msk.ru>
Cc: bugtraq@securityfocus.com, full-disclosure@lists.netsys.com
In-Reply-To: <40B645B4.6060701@msgid.tls.msk.ru>
Message-ID: <Pine.LNX.4.58.0405290510330.12518@twin.jikos.cz>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
On Thu, 27 May 2004, Michael Tokarev wrote:
> I was wrong reading the above code, simple as that.
> Sure, kmalloc(0) will NOT return NULL as I claimed.
> if (size > csizep->cs_size)
> continue;
> Here, when size == 0 (and csizep->cs_size is always > 0),
> the condition is always false, so the next instruction
> will be executed, which is:
> return __kmem_cache_alloc(flags & GFP_DMA ?
> csizep->cs_dmacachep : csizep->cs_cachep, flags);
> which will allocate either 32 or 64 bytes of memory (depending
> on the arch) and return it to the caller.
> So there IS a bug, exactly as described in the original advisory.
> I wonder why noone replied... ;)
Because this all is debate about nothing, as the original advisory was
fake, because you simply can't pass negative optlen to setsockopt()
syscall, so there is nothing to be exploited.
asmlinkage long sys_setsockopt(int fd, int level, int optname, char __user
*optval,
{
int err;
struct socket *sock;
if (optlen < 0)
return -EINVAL;
...
--
JiKos.
