Re: [PATCH 1/2] i2c-dev: don't get i2c adapter via i2c_dev

[Jean Delvare]

Hi Viresh,

A bit of background: at some point in time, the i2c adapter number (as
represented internally by the kernel) could be different from the i2c
device node number (as seen by user-space.) I put an end to this
madness years ago, but it seems some legacy code from that time
survived in i2c-dev.

On Tue,  5 Jul 2016 19:57:06 -0700, Viresh Kumar wrote:
> There is no code protecting i2c_dev to be freed after it is returned
> from i2c_dev_get_by_minor() and using it to access the value which we
> already have (minor) isn't safe really.

I agree that i2c_dev_get_by_minor() looks racy by nature. It is
possible that i2c_dev_get_by_minor() can be removed altogether. There
are 2 other calling locations beyond the one you want to remove. If one
can be removed then I suspect others can be removed as well (maybe with
some more work though.)

If i2c_dev_get_by_minor() needs to stay for whatever reason, then my
next worry is that struct i2c_dev carries an unaccounted reference to
an i2c_adapter. This looks seriously broken. i2c_dev->adap should only
be set on open, and cleared on close.

> Avoid using it and get the adapter directly from 'minor'.
> 
> Signed-off-by: Viresh Kumar <viresh.kumar@xxxxxxxxxx>
> ---
>  drivers/i2c/i2c-dev.c | 7 +------
>  1 file changed, 1 insertion(+), 6 deletions(-)
> 
> diff --git a/drivers/i2c/i2c-dev.c b/drivers/i2c/i2c-dev.c
> index 6ecfd76270f2..66f323fd3982 100644
> --- a/drivers/i2c/i2c-dev.c
> +++ b/drivers/i2c/i2c-dev.c
> @@ -485,13 +485,8 @@ static int i2cdev_open(struct inode *inode, struct file *file)
>  	unsigned int minor = iminor(inode);
>  	struct i2c_client *client;
>  	struct i2c_adapter *adap;
> -	struct i2c_dev *i2c_dev;
> -
> -	i2c_dev = i2c_dev_get_by_minor(minor);
> -	if (!i2c_dev)
> -		return -ENODEV;
>  
> -	adap = i2c_get_adapter(i2c_dev->adap->nr);
> +	adap = i2c_get_adapter(minor);
>  	if (!adap)
>  		return -ENODEV;
>  

This is the most simple fix to your immediate problem. However it
doesn't address the big design issue.

-- 
Jean Delvare
SUSE L3 Support