Re: [PATCH] Version 7 (2.6.23) Smack: Simplified Mandatory Access Control Kernel

[Al Viro <viro@ftp.linux.org.uk>]

On Sun, Oct 14, 2007 at 10:15:42AM -0700, Casey Schaufler wrote:
> This version fixes a major blunder in label handling. The system
> works, but has a serious memory leak that also induces a gradual
> performance degradation. Al Viro gets the credit for pointing out
> that one. Al suggested several other improvements that are not
> included. They should come soon, but I wanted to get this flaw
> out of the code before too many people hit it.

Ahem... This

> +static ssize_t smk_write_doi(struct file *file, const char __user *buf,
> +			     size_t count, loff_t *ppos)
> +{
> +	char temp[80];
> +	int i;
> +
> +	if (!capable(CAP_MAC_OVERRIDE))
> +		return -EPERM;
> +
> +	if (count > sizeof(temp))
> +		return -EINVAL;
> +
> +	if (copy_from_user(temp, buf, count) != 0)
> +		return -EFAULT;
> +
> +	if (sscanf(temp, "%d", &i) != 1)
> +		return -EINVAL;

is not really a missing improvement; it's a geniune undefined behaviour.
temp[] is uninitialized, then you copy there some data that doesn't have
to contain NUL, then you call sscanf().  Boom.  The same goes for the rest
of similar places.

And this

> +static ssize_t smk_read_ambient(struct file *filp, char __user *buf,
> +				size_t cn, loff_t *ppos)
> +{
> +	ssize_t rc;
> +	int asize = strlen(smack_net_ambient) + 1;
> +
> +	if (cn < asize)
> +		return -EINVAL;
> +
> +	if (*ppos != 0)
> +		return 0;
> +
> +	rc = simple_read_from_buffer(buf, cn, ppos, smack_net_ambient, asize);

is honest-to-$DEITY security hole - that file is world-readable and there's
nothing to prevent simple_read_from_buffer() blocking on page-in of buf,
then root writing to that file changing smack_net_ambient and doing kfree()
on the old value - one we'd already passed to simple_read_from_buffer().
At which point reader is about to get whatever data that might land in
whatever that memory gets reused for.

Besides, as I said the last time, smack_net_ambient has every right to
get changed between strlen() and passing argument to simple_read_from_buffer(),
in which case you'll be copying the amount of data that used to be in the
old one, taking it from the new one.  New one might very well be shorter.