Re: [PATCH v2] staging: rtl8712: Move similar execution in to a function.

[Saurav Girepunje]

On 10/09/21 2:30 pm, Dan Carpenter wrote:
> On Fri, Sep 10, 2021 at 01:59:19PM +0530, Saurav Girepunje wrote:
> > In rtl8712_cmd.c function read_macreg_hdl,write_macreg_hdl,write_bbreg_hdl
> > and write_rfreg_hdl all are having same execution.
>
> I get what you're trying to do, because this code is bad and duplicative
> but this is not the right fix.
>
> Let's take read_macreg_hdl() as an example.
>
> Look at how it's called:
>
>     215          switch (pcmd->cmdcode) {
>     216          case GEN_CMD_CODE(_Read_MACREG):
>     217                  read_macreg_hdl(padapter, (u8 *)pcmd);
>     218                  pcmd_r = pcmd;
>     219                  break;
>
> Then look at how it's implemented:
>
>     120  static u8 read_macreg_hdl(struct _adapter *padapter, u8 *pbuf)
>     121  {
>     122          void (*pcmd_callback)(struct _adapter *dev, struct cmd_obj      *pcmd);
>     123          struct cmd_obj *pcmd  = (struct cmd_obj *)pbuf;
>     124
>     125          /*  invoke cmd->callback function */
>     126          pcmd_callback = cmd_callback[pcmd->cmdcode].callback;
>
> So pcmd->cmdcode is GEN_CMD_CODE(_Read_MACREG).  We look that up in the
> cmd_callback[] array and it is:
>
>          {GEN_CMD_CODE(_Read_MACREG), NULL}, /*0*/
>
>     127          if (!pcmd_callback)
>     128                  r8712_free_cmd_obj(pcmd);
>
> So now we no that "pcmd_callback" is NULL meaning it will free "pcmd".
> And if you remember in the caller it does "pcmd_r = pcmd;" but "pcmd"
> is freed so that's going to lead to a use after free in r8712_cmd_thread().
> It's garbage and the patch doesn't really help.

One more thought here after the

    127          if (!pcmd_callback)
    128                  r8712_free_cmd_obj(pcmd);

r8712_free_cmd_obj(pcmd); we could do  pcmd = NULL; so in the caller 
when it will do "pcmd_r = pcmd;" it is actually making NULL to pcmd_r. 
On r8712_cmd_thread there is check for pcmd is NULL or not before 
execution on pcmd.

pcmd = cmd_hdl_filter(padapter, pcmd);
	if (pcmd) { /* if pcmd != NULL, cmd will be handled by f/w */

Please let me know you thought on this dan.

> The right way to fix it is to get rid of the cmd_callback[] array.
>
>     129          else
>     130                  pcmd_callback(padapter, pcmd);
>     131          return H2C_SUCCESS;
>     132  }
>
> Getting rid of the cmd_callback[] array looks like this.  In
> read_rfreg_hdl() we know that the callback is r8712_getbbrfreg_cmdrsp_callback()
> so we can call that directly.
>
> diff --git a/drivers/staging/rtl8712/rtl8712_cmd.c b/drivers/staging/rtl8712/rtl8712_cmd.c
> index e9294e1ed06e..51a6abb27d41 100644
> --- a/drivers/staging/rtl8712/rtl8712_cmd.c
> +++ b/drivers/staging/rtl8712/rtl8712_cmd.c
> @@ -174,11 +174,7 @@ static u8 read_rfreg_hdl(struct _adapter *padapter, u8 *pbuf)
>   
>   	if (pcmd->rsp && pcmd->rspsz > 0)
>   		memcpy(pcmd->rsp, (u8 *)&val, pcmd->rspsz);
> -	pcmd_callback = cmd_callback[pcmd->cmdcode].callback;
> -	if (!pcmd_callback)
> -		r8712_free_cmd_obj(pcmd);
> -	else
> -		pcmd_callback(padapter, pcmd);
> +	r8712_getbbrfreg_cmdrsp_callback(padapter, pcmd);
>   	return H2C_SUCCESS;
>   }
>   
>
> regards,
> dan carpenter

Thanks Dan for review.

Regards,
Saurav Girepunje