On Mon, May 13, 2019 at 05:37:41PM +0800, Gen Zhang wrote:
> On Mon, May 13, 2019 at 09:36:19AM +0200, Greg KH wrote:
> > > Signed-off-by: Gen Zhang <blackgod016...@gmail.com>
> > > ---
> > > --- drivers/tty/vt/vt.c
> > > +++ drivers/tty/vt/vt.c
> > > @@ -3349,10 +3349,14 @@ static int __init con_init(void)
> > >
> > > for (currcons = 0; currcons < MIN_NR_CONSOLES; currcons++) {
> > > vc_cons[currcons].d = vc = kzalloc(sizeof(struct vc_data),
> > > GFP_NOWAIT);
> > > + if (!vc_cons[currcons].d || !vc)
> > > + goto err_vc;
> >
> > What about the other memory that was allocated? You never free that.
> >
> > > INIT_WORK(&vc_cons[currcons].SAK_work, vc_SAK);
> > > tty_port_init(&vc->port);
> > > visual_init(vc, currcons, 1);
> > > vc->vc_screenbuf = kzalloc(vc->vc_screenbuf_size, GFP_NOWAIT);
> > > + if (!vc->vc_screenbuf)
> > > + goto err_vc_screenbuf;
> >
> > Same here, you are now leaking memory.
> >
> > Did you test this patch out with a kmalloc function that can fail? If
> > not, please try to do so.
> >
> > thanks,
> >
> > greg k-h
> Hi, Greg
> 1. I re-examined the source code.
> For vc_cons[currcons].d and vc allocation fail, we may need to free
> vc->vc_screenbuf from the previous loop. So kfree(vc->vc_screenbuf)
> need to be added to err_vc;
> As for vc->vc_screenbuf allocation fail, I don't think there is other
> memory need to be freed. Because in function con_init, there's no other
> allocation operations except this two kzalloc functions. And in
> err_vc_screenbuf, vc_cons[currcons].d and vc is freed in the patch.
You have to unwind the loop and free and uninitialize all of the other
things you just created as well.
> 2. I tried to test this patch with a compiled kernel in QEMU but
> failed. Testing this is out of my skills. So is there any other ways
> to test this patch?
qemu should work just fine, I don't know what else to suggest. Run it
on "real hardware" with a kmalloc function modified to fail this
allocation?
good luck!
greg k-h
