Re: [patch v3 1/3] drivers: jtag: Add JTAG core driver

From: Arnd Bergmann
Date: Tue Aug 15 2017 - 07:16:14 EST


On Tue, Aug 15, 2017 at 12:00 PM, Oleksandr Shamray
<oleksandrs@xxxxxxxxxxxx> wrote:

> + case JTAG_IOCXFER:
> + if (copy_from_user(&xfer, varg, sizeof(struct jtag_xfer)))
> + return -EFAULT;
> +
> + if (xfer.length >= JTAG_MAX_XFER_DATA_LEN)
> + return -EFAULT;
> +
> + user_tdio_data = xfer.tdio;
> + xfer.tdio = jtag_copy_from_user((void __user *)user_tdio_data,
> + xfer.length);
> + if (!xfer.tdio)
> + return -ENOMEM;

This is not safe for 32-bit processes on 64-bit kernels, since the
structure layout differs for the pointer member. It's better to always
use either rework the structure to avoid the pointer, or to use a
__u64 member to store it, and then use u64_to_user_ptr()
to convert it in the kernel.

> + case JTAG_GIOCSTATUS:
> + if (jtag->ops->status_get)
> + err = jtag->ops->status_get(jtag,
> + (enum jtag_endstate *)&value);

This pointer cast is also not safe, as an enum might have a different
size than the 'value' variable that is not an enum. I think you need to
change the argument type for the callback, or maybe use another
intermediate.

> +static int jtag_open(struct inode *inode, struct file *file)
> +{
> + struct jtag *jtag = container_of(inode->i_cdev, struct jtag, cdev);
> +
> + spin_lock(&jtag->lock);
> +
> + if (jtag->is_open) {
> + dev_info(NULL, "jtag already opened\n");
> + spin_unlock(&jtag->lock);
> + return -EBUSY;
> + }
> +
> + jtag->is_open = true;
> + file->private_data = jtag;
> + spin_unlock(&jtag->lock);
> + return 0;
> +}

Does the 'is_open' flag protect you from something that doesn't
also happen after a 'dup()' call on the file descriptor?

> + * @mode: access mode
> + * @type: transfer type
> + * @direction: xfer direction
> + * @length: xfer bits len
> + * @tdio : xfer data array
> + * @endir: xfer end state
> + *
> + * Structure represents interface to Aspeed JTAG device for jtag sdr xfer
> + * execution.
> + */
> +struct jtag_xfer {
> + __u8 mode;
> + __u8 type;
> + __u8 direction;
> + __u32 length;
> + __u8 *tdio;
> + __u8 endstate;
> +};

As mentioned above, the pointer in here makes it incompatible. Also,
you should reorder the members to avoid the implied padding.
Moving the 'endstate' before 'length' is sufficient.

Arnd