Re: [REGRESSION][BISECTED] flood of "hid-generic ... control queue full" since v5.14-rc1

[Alan Stern]

On Mon, Aug 16, 2021 at 04:13:47PM +0200, Michal Kubecek wrote:
> On Mon, Aug 16, 2021 at 03:26:20PM +0200, Greg Kroah-Hartman wrote:
> > On Mon, Aug 16, 2021 at 03:00:59PM +0200, Michal Kubecek wrote:
> > > Hello,
> > > 
> > > starting with v5.14-rc1, my kernel log gets flooded with messages
> > > 
> > >   hid-generic 0003:051D:0002.0002: control queue full
> > > 
> > > at rate of ~33 per second. Device 051d:0002 is an APC UPS (BR-650 VA).
> > > I bisected the issue to commit
> > > 
> > >   7652dd2c5cb7 ("USB: core: Check buffer length matches wLength for control transfers")
> > > 
> > > Reverting this commit on top of v5.14-rc6 resolves the issue. I suspect
> > > the problem is some missing cleanup when usb_submit_urb() bails out on
> > > the newly added check but I'm not familiar enough with the code to see
> > > what is missing or if the problem is on USB or HID side.

...

> Looking at the code, the primary problem seems to be that the "else"
> branch in hid_submit_ctrl() recalculates transfer_buffer_length to
> a rounded up value but assigns the original length to wLength.

Looks like you found the bug.  Fixing it might be as simple as setting 
len = padlen in that "else" branch.  You could then combine the 
transfer_buffer_length assignment with the one in the "if" branch and 
hoist them out after the entire "if" statement.

Alan Stern