Set logo_shown to FBCON_LOGO_CANSHOW when the vc was deallocated.
syzkaller report: https://lkml.org/lkml/2020/3/27/403
general protection fault, probably for non-canonical address
0xdffffc000000006c: 0000 [#1] SMP KASAN
KASAN: null-ptr-deref in range [0x0000000000000360-0x0000000000000367]
RIP: 0010:fbcon_switch+0x28f/0x1740
drivers/video/fbdev/core/fbcon.c:2260
Call Trace:
redraw_screen+0x2a8/0x770 drivers/tty/vt/vt.c:1008
vc_do_resize+0xfe7/0x1360 drivers/tty/vt/vt.c:1295
fbcon_init+0x1221/0x1ab0 drivers/video/fbdev/core/fbcon.c:1219
visual_init+0x305/0x5c0 drivers/tty/vt/vt.c:1062
do_bind_con_driver+0x536/0x890 drivers/tty/vt/vt.c:3542
do_take_over_console+0x453/0x5b0 drivers/tty/vt/vt.c:4122
do_fbcon_takeover+0x10b/0x210 drivers/video/fbdev/core/fbcon.c:588
fbcon_fb_registered+0x26b/0x340 drivers/video/fbdev/core/fbcon.c:3259
do_register_framebuffer drivers/video/fbdev/core/fbmem.c:1664 [inline]
register_framebuffer+0x56e/0x980 drivers/video/fbdev/core/fbmem.c:1832
dlfb_usb_probe.cold+0x1743/0x1ba3 drivers/video/fbdev/udlfb.c:1735
usb_probe_interface+0x310/0x800 drivers/usb/core/driver.c:374
accessing vc_cons[logo_shown].d->vc_top causes the bug.
Reported-by: [email protected]
Signed-off-by: Qiujun Huang <[email protected]>
---
drivers/video/fbdev/core/fbcon.c | 3 +++
1 file changed, 3 insertions(+)
diff --git a/drivers/video/fbdev/core/fbcon.c b/drivers/video/fbdev/core/fbcon.c
index bb6ae995c2e5..5eb3fc90f9f6 100644
--- a/drivers/video/fbdev/core/fbcon.c
+++ b/drivers/video/fbdev/core/fbcon.c
@@ -1283,6 +1283,9 @@ static void fbcon_deinit(struct vc_data *vc)
if (!con_is_bound(&fb_con))
fbcon_exit();
+ if (vc->vc_num == logo_shown)
+ logo_shown = FBCON_LOGO_CANSHOW;
+
return;
}
--
2.17.1
Hi Qiujun
On Sun, Mar 29, 2020 at 04:56:47PM +0800, Qiujun Huang wrote:
> Set logo_shown to FBCON_LOGO_CANSHOW when the vc was deallocated.
>
> syzkaller report: https://lkml.org/lkml/2020/3/27/403
> general protection fault, probably for non-canonical address
> 0xdffffc000000006c: 0000 [#1] SMP KASAN
> KASAN: null-ptr-deref in range [0x0000000000000360-0x0000000000000367]
> RIP: 0010:fbcon_switch+0x28f/0x1740
> drivers/video/fbdev/core/fbcon.c:2260
>
> Call Trace:
> redraw_screen+0x2a8/0x770 drivers/tty/vt/vt.c:1008
> vc_do_resize+0xfe7/0x1360 drivers/tty/vt/vt.c:1295
> fbcon_init+0x1221/0x1ab0 drivers/video/fbdev/core/fbcon.c:1219
> visual_init+0x305/0x5c0 drivers/tty/vt/vt.c:1062
> do_bind_con_driver+0x536/0x890 drivers/tty/vt/vt.c:3542
> do_take_over_console+0x453/0x5b0 drivers/tty/vt/vt.c:4122
> do_fbcon_takeover+0x10b/0x210 drivers/video/fbdev/core/fbcon.c:588
> fbcon_fb_registered+0x26b/0x340 drivers/video/fbdev/core/fbcon.c:3259
> do_register_framebuffer drivers/video/fbdev/core/fbmem.c:1664 [inline]
> register_framebuffer+0x56e/0x980 drivers/video/fbdev/core/fbmem.c:1832
> dlfb_usb_probe.cold+0x1743/0x1ba3 drivers/video/fbdev/udlfb.c:1735
> usb_probe_interface+0x310/0x800 drivers/usb/core/driver.c:374
>
> accessing vc_cons[logo_shown].d->vc_top causes the bug.
>
> Reported-by: [email protected]
> Signed-off-by: Qiujun Huang <[email protected]>
> ---
> drivers/video/fbdev/core/fbcon.c | 3 +++
> 1 file changed, 3 insertions(+)
>
> diff --git a/drivers/video/fbdev/core/fbcon.c b/drivers/video/fbdev/core/fbcon.c
> index bb6ae995c2e5..5eb3fc90f9f6 100644
> --- a/drivers/video/fbdev/core/fbcon.c
> +++ b/drivers/video/fbdev/core/fbcon.c
> @@ -1283,6 +1283,9 @@ static void fbcon_deinit(struct vc_data *vc)
> if (!con_is_bound(&fb_con))
> fbcon_exit();
>
> + if (vc->vc_num == logo_shown)
> + logo_shown = FBCON_LOGO_CANSHOW;
> +
> return;
> }
Looks much better than the previous version.
Acked-by: Sam Ravnborg <[email protected]>
I expect Bartlomiej to review/apply.
Sam
On Mon, Mar 30, 2020 at 09:16:19PM +0200, Sam Ravnborg wrote:
> Hi Qiujun
>
> On Sun, Mar 29, 2020 at 04:56:47PM +0800, Qiujun Huang wrote:
> > Set logo_shown to FBCON_LOGO_CANSHOW when the vc was deallocated.
> >
> > syzkaller report: https://lkml.org/lkml/2020/3/27/403
> > general protection fault, probably for non-canonical address
> > 0xdffffc000000006c: 0000 [#1] SMP KASAN
> > KASAN: null-ptr-deref in range [0x0000000000000360-0x0000000000000367]
> > RIP: 0010:fbcon_switch+0x28f/0x1740
> > drivers/video/fbdev/core/fbcon.c:2260
> >
> > Call Trace:
> > redraw_screen+0x2a8/0x770 drivers/tty/vt/vt.c:1008
> > vc_do_resize+0xfe7/0x1360 drivers/tty/vt/vt.c:1295
> > fbcon_init+0x1221/0x1ab0 drivers/video/fbdev/core/fbcon.c:1219
> > visual_init+0x305/0x5c0 drivers/tty/vt/vt.c:1062
> > do_bind_con_driver+0x536/0x890 drivers/tty/vt/vt.c:3542
> > do_take_over_console+0x453/0x5b0 drivers/tty/vt/vt.c:4122
> > do_fbcon_takeover+0x10b/0x210 drivers/video/fbdev/core/fbcon.c:588
> > fbcon_fb_registered+0x26b/0x340 drivers/video/fbdev/core/fbcon.c:3259
> > do_register_framebuffer drivers/video/fbdev/core/fbmem.c:1664 [inline]
> > register_framebuffer+0x56e/0x980 drivers/video/fbdev/core/fbmem.c:1832
> > dlfb_usb_probe.cold+0x1743/0x1ba3 drivers/video/fbdev/udlfb.c:1735
> > usb_probe_interface+0x310/0x800 drivers/usb/core/driver.c:374
> >
> > accessing vc_cons[logo_shown].d->vc_top causes the bug.
> >
> > Reported-by: [email protected]
> > Signed-off-by: Qiujun Huang <[email protected]>
> > ---
> > drivers/video/fbdev/core/fbcon.c | 3 +++
> > 1 file changed, 3 insertions(+)
> >
> > diff --git a/drivers/video/fbdev/core/fbcon.c b/drivers/video/fbdev/core/fbcon.c
> > index bb6ae995c2e5..5eb3fc90f9f6 100644
> > --- a/drivers/video/fbdev/core/fbcon.c
> > +++ b/drivers/video/fbdev/core/fbcon.c
> > @@ -1283,6 +1283,9 @@ static void fbcon_deinit(struct vc_data *vc)
> > if (!con_is_bound(&fb_con))
> > fbcon_exit();
> >
> > + if (vc->vc_num == logo_shown)
> > + logo_shown = FBCON_LOGO_CANSHOW;
> > +
> > return;
> > }
>
> Looks much better than the previous version.
> Acked-by: Sam Ravnborg <[email protected]>
>
> I expect Bartlomiej to review/apply.
Especially for bugfixes I think better to push quicker than wait for
others to ... the point with drm-misc is real group maintainership and
benefitting from the flexibility, not reflecting the same strict hierarchy
but in a flat tree to make it look like it doesn't exist :-)
Applied to drm-misc-next-fixes with a cc: stable.
-Daniel
--
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch