lkml.org 
[lkml]   [2023]   [Jan]   [6]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH] tty: vt: add some NULL checks for vc_data
On Tue, Jan 03, 2023 at 10:01:15PM -0500, Hang Zhang wrote:
> On Tue, Jan 3, 2023 at 4:24 AM Jiri Slaby <jirislaby@kernel.org> wrote:
> >
> > On 29. 12. 22, 7:41, Hang Zhang wrote:
> > > vc_selection(), do_blank_screen() and scrollfront() all access "vc_data"
> > > structures obtained from the global "vc_cons[fg_console].d", which can
> > > be freed and nullified (e.g., in the error path of vc_allocate()). But
> > > these functions don't have any NULL checks against the pointers before
> > > dereferencing them, causing potentially use-after-free or null pointer
> > > dereference.
> >
> > Could you elaborate under what circumstances is fg_console set to a
> > non-allocated console?
>
> Hi, Jiri, thank you for your reply! I am not a developer for tty
> subsystem, so the reasoning here is based on my best-effort code
> reading. Please correct me if I am wrong.
>
> This patch is based on several observations:
>
> (1) at the beginning of vc_selection() (where one NULL check is
> inserted in this patch), poke_blanked_console() is invoked, which
> explicitly checks whether "vc_cons[fg_console].d" is NULL, suggesting
> the possibility of "fg_console" associated with an unallocated console
> at this point. However, poke_blanked_console() returns "void", so
> even if "fg_console" is NULL, after returning to vc_selection(),
> it will just keep executing, resulting in the possible NULL pointer
> dereference later ("vc" in vc_selection() can be "vc_cons[fg_console].d"
> if called from set_selection_kernel()). So this patch actually tries
> to make the already existing NULL check take effect on the control
> flow (e.g., early return if NULL).

But again, how can that value ever be NULL?

And why are you returning "success" if it is?

> (2) a similar NULL check for "vc_cons[fg_console].d" can also be found
> in do_unblank_screen() ("if (!vc_cons_allocated(fg_console))") before
> accessing the corresponding "vc_data". I do notice that the NULL check
> has a comment "/* impossible */", but the check has not been removed so
> far. My guess is that there might still be a chance that it can be
> unallocated at that point.

Please verify that this really ever could be NULL or not.

thanks,

greg k-h

\
 
 \ /
  Last update: 2023-03-26 23:29    [W:0.632 / U:0.436 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site