Re: [PATCH riscv64] kobject: fix WARNING in input_register_device

From: Edward Adam Davis <eadavis@qq.com>
To: gregkh@linuxfoundation.org
Cc: eadavis@qq.com, linux-input@vger.kernel.org,
	linux-kernel@vger.kernel.org, linux-usb@vger.kernel.org,
	rafael@kernel.org,
	syzbot+8e41bb0c055b209ebbf4@syzkaller.appspotmail.com,
	syzkaller-bugs@googlegroups.com
Subject: Re: [PATCH riscv64] kobject: fix WARNING in input_register_device
Date: Tue, 13 Feb 2024 08:43:26 +0800	[thread overview]
Message-ID: <tencent_DDCFB377C3642974A3A3A44D176B776DA605@qq.com> (raw)
In-Reply-To: <2024020836-flypaper-relapse-5c97@gregkh>

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@syzkaller.appspotmail.com
> > > > Signed-off-by: Edward Adam Davis <eadavis@qq.com>
> > > > ---
> > > >  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.

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;                                                                                                                                                                        
  1         }

I believe that this issue was introduced by:
7eff2e7a8b65 - Driver core: change add_ueventvar to use a struct.

thanks,
edward.