Re: [RFC PATCH net-next 9/9] ethtool: Add ability to flash transceiver modules' firmware

[Jakub Kicinski]

On Tue, 23 Jan 2024 13:05:16 +0000 Danielle Ratson wrote:
> > GENL_REQ_ATTR_CHECK, and you can check it in the caller, before taking
> > rtnl_lock.
> >   
> 
> OK, np. The idea was to have module_flash_fw() that checks the attrs
> and extract them into params and ethnl_act_module_fw_flash() should
> be free from those checks. But if so, maybe this separation is
> redundant and should combine the two?

No strong preference, whatever looks better :)
To use GENL_REQ_ATTR_CHECK() I think you'll need to pass genl_info here.
You can either to that or move the validation.

> > > +  
> > tb[ETHTOOL_A_MODULE_FW_FLASH_FILE_NAME],  
> > > +				    "File name attribute is missing");
> > > +		return -EINVAL;
> > > +	}
> > > +
> > > +	params.file_name =
> > > +		nla_data(tb[ETHTOOL_A_MODULE_FW_FLASH_FILE_NAME]);  
> > 
> > Hm. I think you copy the param struct by value to the work container.
> > nla_data() is in the skb which is going to get freed after _ACT returns.
> > So if anyone tries to access the name from the work it's going to UAF?  
> 
> The file_name parameter is not really needed inside the work. Once we
> called request_firmware_direct(), we have all that we need in
> module_fw->fw. Do we still need to avoid that situation? If so, can
> you please suggest how?

I'd pass it to module_flash_fw_schedule() as a separate argument, if it
doesn't have to be saved.