Re: [PATCH] loop: mask loop_control_ioctl parameter only as minor

From: wangyangbo
Date: Thu Nov 18 2021 - 20:50:29 EST




> On Nov 18, 2021, at 22:15, Tetsuo Handa <penguin-kernel@xxxxxxxxxxxxxxxxxxx> wrote:
>
> On 2021/11/18 11:36, wangyangbo wrote:
>> @@ -2170,11 +2170,11 @@ static long loop_control_ioctl(struct file *file, unsigned int cmd,
>> {
>> switch (cmd) {
>> case LOOP_CTL_ADD:
>> - return loop_add(parm);
>> + return loop_add(MINOR(parm));
>
> Better to return -EINVAL or something if out of minor range?
Definitely, EINVAL or EDOM, which do you think is better?

>
>> case LOOP_CTL_REMOVE:
>> - return loop_control_remove(parm);
>> + return loop_control_remove(MINOR(parm));
>
> This is bad, for this change makes
>
> if (idx < 0) {
> pr_warn("deleting an unspecified loop device is not supported.\n");
> return -EINVAL;
> }
>
> dead code by masking the argument to 0-1048575 range.

But ioctl param is unsigned long, I think this need to sanitize.

>> case LOOP_CTL_GET_FREE:
>> - return loop_control_get_free(parm);
>> + return loop_control_get_free(MINOR(parm));
>
> This is pointless, for the passed argument is not used.
> By the way, didn't someone already propose removal of this argument?

I don't find this in mail list, but I would like to sanitize that code.