2022-02-22 05:17:43

by Xiaomeng Tong

Subject: [PATCH] vt_ioctl: fix potential spectre v1 in VT_DISALLOCATE

In VT_ACTIVATE an almost identical code path has been patched
with array_index_nospec. In the VT_DISALLOCATE path, the arg is
the user input from a system call argument and lately used as a index
for vc_cons[index].d access, which can be reached through path like
vt_disallocate->vc_busy or vt_disallocate->vc_deallocate.
For consistency both code paths should have the same mitigations
applied.

Signed-off-by: Xiaomeng Tong <[email protected]>
---
drivers/tty/vt/vt_ioctl.c | 9 ++++++---
1 file changed, 6 insertions(+), 3 deletions(-)

diff --git a/drivers/tty/vt/vt_ioctl.c b/drivers/tty/vt/vt_ioctl.c
index 580136986..acd53c1c9 100644
--- a/drivers/tty/vt/vt_ioctl.c
+++ b/drivers/tty/vt/vt_ioctl.c
@@ -898,10 +898,13 @@ int vt_ioctl(struct tty_struct *tty,
if (arg > MAX_NR_CONSOLES)
return -ENXIO;

- if (arg == 0)
+ if (arg == 0) {
vt_disallocate_all();
- else
- return vt_disallocate(--arg);
+ } else {
+ --arg;
+ arg = array_index_nospec(arg, MAX_NR_CONSOLES);
+ return vt_disallocate(arg);
+ }
break;

case VT_RESIZE:
--
2.17.1

2022-02-22 06:56:25

by Jiri Slaby

Subject: Re: [PATCH] vt_ioctl: fix potential spectre v1 in VT_DISALLOCATE

On 22. 02. 22, 4:56, Xiaomeng Tong wrote:
> In VT_ACTIVATE an almost identical code path has been patched
> with array_index_nospec. In the VT_DISALLOCATE path, the arg is
> the user input from a system call argument and lately used as a index
> for vc_cons[index].d access, which can be reached through path like
> vt_disallocate->vc_busy or vt_disallocate->vc_deallocate.
> For consistency both code paths should have the same mitigations
> applied.
>
> Signed-off-by: Xiaomeng Tong <[email protected]>
> ---
> drivers/tty/vt/vt_ioctl.c | 9 ++++++---
> 1 file changed, 6 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/tty/vt/vt_ioctl.c b/drivers/tty/vt/vt_ioctl.c
> index 580136986..acd53c1c9 100644
> --- a/drivers/tty/vt/vt_ioctl.c
> +++ b/drivers/tty/vt/vt_ioctl.c
> @@ -898,10 +898,13 @@ int vt_ioctl(struct tty_struct *tty,
> if (arg > MAX_NR_CONSOLES)
> return -ENXIO;
>
> - if (arg == 0)
> + if (arg == 0) {
> vt_disallocate_all();

Could you add a break; here and remove the else branching completely?

> - else
> - return vt_disallocate(--arg);
> + } else {
> + --arg;
> + arg = array_index_nospec(arg, MAX_NR_CONSOLES);
You could simply do:
arg = array_index_nospec(arg - 1, MAX_NR_CONSOLES);
and remove
--arg;
completely.

> + return vt_disallocate(arg);
> + }
> break;

When you do the above, this break is superfluous.

thanks,
--
js
suse labs

2022-02-24 00:47:41

by Xiaomeng Tong

Subject: Re: [PATCH] vt_ioctl: fix potential spectre v1 in VT_DISALLOCATE

On Tue, 22 Feb 2022 07:56:13 +0100, Jiri Slaby wrote:
> Could you add a break; here and remove the else branching completely?
> ...
> You could simply do:
> arg = array_index_nospec(arg - 1, MAX_NR_CONSOLES);
> and remove
> --arg;
> completely.

Already adjusted as suggested in Patch v2, thanks!

Best regards,
--
Xiaomeng Tong