Re: [Patch] Off by one in drivers/usb/serial/usb-serial.c

[Greg KH]

On Mon, Jun 26, 2006 at 12:59:21AM +0200, Eric Sesterhenn / Snakebyte wrote:
> * Mikael Pettersson (mikpe@xxxxxxxx) wrote:
> > On Wed, 21 Jun 2006 23:28:17 +0200, Eric Sesterhenn wrote:
> > > this fixes coverity id #554. since serial table
> > > is defines as serial_table[SERIAL_TTY_MINORS] we
> > > should make sure we dont acess with an index
> > > of SERIAL_TTY_MINORS.
> > > 
> > > Signed-off-by: Eric Sesterhenn <snakebyte@xxxxxx>
> > > 
> > > --- linux-2.6.17-git2/drivers/usb/serial/usb-serial.c.orig	2006-06-21 23:24:07.000000000 +0200
> > > +++ linux-2.6.17-git2/drivers/usb/serial/usb-serial.c	2006-06-21 23:25:12.000000000 +0200
> > > @@ -83,7 +83,7 @@ static struct usb_serial *get_free_seria
> > >  
> > >  		good_spot = 1;
> > >  		for (j = 1; j <= num_ports-1; ++j)
> > > -			if ((i+j >= SERIAL_TTY_MINORS) || (serial_table[i+j])) {
> > > +			if ((i+j >= SERIAL_TTY_MINORS-1)||(serial_table[i+j])) {
> > >  				good_spot = 0;
> > >  				i += j;
> > >  				break;
> > 
> > Where is the access coverity complained about? If it's the serial_table[i+j]
> > quoted above, then the original code is OK since i+j < SERIAL_TTY_MINORS is
> > an invariant in that subexpression.
> > 
> > And the other accesses to serial_table[] in get_free_serial() are also only
> > done when the index is < SERIAL_TTY_MINORS.
> 
> guess i was too quick on that one, sorry. Here is the coverity
> report for completeness.
> 
> Event assignment: Assigning "1" to "j"
> Also see events: [overrun-local]
> At conditional (11): "j <= (num_ports - 1)" taking true path
> At conditional (16): "j <= (num_ports - 1)" taking true path
> 
> 85   			for (j = 1; j <= num_ports-1; ++j)
> 
> Event overrun-local: Overrun of static array "serial_table" of size 255
> at position 255 with index variable "(i + j)"
> Also see events: [assignment]
> At conditional (12): "(i + j) >= 255" taking true path
> At conditional (17): "(i + j) >= 255" taking false path
> 
> 86   				if ((i+j >= SERIAL_TTY_MINORS) ||
> (serial_table[i+j])) {
> 87   					good_spot = 0;
> 88   					i += j;
> 89   					break;
> 90   				}

So, what does this mean?  That coverity is broken, yet again?

I'm getting very tired of these false positives from them, it is getting
so that I can't trust the output of the tool at all :(

thanks,

greg k-h
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/