On Tue, 13 Feb 2024 08:20:52 +0100, Greg KH wrote:
> On Thu, 8 Feb 2024 10:56:00, Greg KH wrote:
> On Tue, Feb 13, 2024 at 08:43:26AM +0800, Edward Adam Davis wrote:
> > On Thu, 8 Feb 2024 12:25:10 +0000, Greg KH wrote:
> > > On Thu, Feb 08, 2024 at 07:37:56PM +0800, Edward Adam Davis wrote:
> > > > On Thu, 8 Feb 2024 10:56:00, Greg KH wrote:
> > > > > > The input_add_uevent_modalias_var()->input_print_modalias() will add 1684 bytes
> > > > > > of data to env, which will result in insufficient memory allocated to the buf
> > > > > > members of env.
> > > > >
> > > > > What is "env"? And can you wrap your lines at 72 columns please?
> > > > env is an instance of struct kobj_uevent_env.
> > >
> > > Ok, be specific please in your changelog text, otherwise we can't really
> > > understand what is happening.
> > >
> > > > > > Reported-and-tested-by: syzbot+8e41bb0c055b209ebbf4@xxxxxxxxxxxxxxxxxxxxxxxxx
> > > > > > Signed-off-by: Edward Adam Davis <eadavis@xxxxxx>
> > > > > > ---
> > > > > > include/linux/kobject.h | 2 +-
> > > > > > 1 file changed, 1 insertion(+), 1 deletion(-)
> > > > > >
> > > > > > diff --git a/include/linux/kobject.h b/include/linux/kobject.h
> > > > > > index c30affcc43b4..74b37b6459cd 100644
> > > > > > --- a/include/linux/kobject.h
> > > > > > +++ b/include/linux/kobject.h
> > > > > > @@ -30,7 +30,7 @@
> > > > > >
> > > > > > #define UEVENT_HELPER_PATH_LEN 256
> > > > > > #define UEVENT_NUM_ENVP 64 /* number of env pointers */
> > > > > > -#define UEVENT_BUFFER_SIZE 2048 /* buffer for the variables */
> > > > > > +#define UEVENT_BUFFER_SIZE 2560 /* buffer for the variables */
> > > > >
> > > > > That's an odd number, why that? Why not just a page? What happens if
> > > > > some other path wants more?
> > > > An increase of 512 bytes is sufficient for the current issue. Do not consider
> > > > the problem of hypothetical existence.
> > >
> > > Why is this 512 bytes sufficient now? What changed to cause this?
> > There is the following code in input_print_modalias():
> >
> > drivers/input/input.c
> > 1 len += input_print_modalias_bits(buf + len, size - len,
> > 1403 'k', id->keybit, KEY_MIN_INTERESTING, KEY_MAX);
> > This code will add up to 2608 bytes of data to env at most.
> > (KEY_MAX - KEY_MIN_INTERESTING) * 4 = (256 * 3 - 1 - 113 ) * 4 = (765 - 113) * 4 = 652 * 4 = 2608 bytes。
> > Note: In the expression, 4 represents 3 bytes of hexadecimal data and 1 byte of comma.
>
> So your change above is wrong and will not work for the max size?
Yes.
>
> Why not restrict the modalias here to fit instead of overflowing? Odds
> are we should be checking this properly no matter what the value is
> changed to, right?
Right.
It may be necessary to deepen our understanding of this piece of code before
fixing this issue internally.
>
> > include/uapi/linux/input-event-codes.h
> > 188 #define KEY_MUTE 113
> > 807 #define KEY_MIN_INTERESTING KEY_MUTE
> > 808 #define KEY_MAX 0x2ff
> > During my actual testing process, I found that a total of 1684 bytes were
> > contributed in input_print_modalias().
> > >
> > > And how can we detect this automatically in the future? Shouldn't we
> > > just be truncating the buffer instead of having an overflow?
> > >
> > > > > And what's causing the input stack to have so many variables all of a
> > > > > sudden, what changed to cause this? Is this a bugfix for a specific
> > > > > commit that needs to be backported to older kernels? Why did this
> > > > > buffer size all of a sudden be too small?
> > > > The result of my analysis is that several members of struct input_dev are too
> > > > large, such as its member keybit.
> > >
> > > And when did that change? What commit id? What prevents it from
> > > growing again and us needing to change this again?
> > The code that caused this issue has been introduced for a long time, and it is
> > speculated that it was due to the fact that the warning in add_uevent_var() was
> > returned directly to ENOMEM without being taken seriously.
> >
> > lib/kobject_uevent.c
> > 2 if (len >= (sizeof(env->buf) - env->buflen)) {
> > 1 WARN(1, KERN_ERR "add_uevent_var: buffer size too small\n");
> > 672 return -ENOMEM;
>
> Odd line numbers?
>
> Anyway, we should get rid of the WARN() as that will cause crashes, and
> just handle it properly there.
>
>
> > 1 }
> >
> > I believe that this issue was introduced by:
> > 7eff2e7a8b65 - Driver core: change add_ueventvar to use a struct.
>
> In 2007? And never been actually hit since then? So is this a real
> issue? :)
Yes. But as I mentioned earlier, in add_uevent_var(), it will exit directly after
a warning, so this issue has not been given enough attention, perhaps it has
happened many times.
thanks,
edward.
