Re: [PATCH] tpm: fix a race condition tpm2_unseal_trusted()

From: Jason Gunthorpe <jgunthorpe@obsidianresearch.com>
To: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>
Cc: Peter Huewe <peterhuewe@gmx.de>,
	linux-security-module@vger.kernel.org, stable@vger.kernel.org,
	Marcel Selhorst <tpmdd@selhorst.net>,
	"moderated list:TPM DEVICE DRIVER" 
	<tpmdd-devel@lists.sourceforge.net>,
	open list <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH] tpm: fix a race condition tpm2_unseal_trusted()
Date: Thu, 25 Aug 2016 12:30:59 -0600	[thread overview]
Message-ID: <20160825183059.GB1142@obsidianresearch.com> (raw)
In-Reply-To: <1472000243-7088-1-git-send-email-jarkko.sakkinen@linux.intel.com>

On Tue, Aug 23, 2016 at 08:57:22PM -0400, Jarkko Sakkinen wrote:

> +     if (flags & TPM_TRANSMIT_LOCK)
> +             mutex_lock(&chip->tpm_mutex);

I think I would invert this. UNLOCKED is the exceptional case, so I'd
make the 0 flags lock. If we see UNLOCKED in the caller then we know
to audit for locking, 0 is much less obvious.

> @@ -576,7 +576,7 @@ static int tpm2_load(struct tpm_chip *chip,
>  		goto out;
>  	}
>  
> -	rc = tpm_transmit_cmd(chip, buf.data, PAGE_SIZE, "loading blob");
> +	rc = __tpm_transmit_cmd(chip, buf.data, PAGE_SIZE, "loading blob", 0);

All these points should accept a flags too and the caller should pass
in the TPM_TRASNMIT_UNLOCKED if it needs it..

> +	mutex_lock(&chip->tpm_mutex);
>  	rc = tpm2_load(chip, payload, options, &blob_handle);
>  	if (rc)

So when we read here we see the pattern:

> +	mutex_lock(&chip->tpm_mutex);
>  	rc = tpm2_load(chip, payload, options, &blob_handle, TPM_TRASNMIT_UNLOCKED);

Which is much easier to audit..

Jason