RE: [PATCH] driver: input: fix UBSAN warning in input_defuzz_abs_event

[liujian (CE)]

Best Regards,
liujian

> -----Original Message-----
> From: Dmitry Torokhov [mailto:dmitry.torokhov@xxxxxxxxx]
> Sent: Tuesday, November 13, 2018 3:49 AM
> To: liujian (CE) <liujian56@xxxxxxxxxx>
> Cc: linux-input@xxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx
> Subject: Re: [PATCH] driver: input: fix UBSAN warning in
> input_defuzz_abs_event
> 
> Hi,
> 
> On Fri, Nov 02, 2018 at 09:48:51PM +0800, liujian wrote:
> > syzkaller triggered a UBCAN warning:
> >
> > [  196.188950] UBSAN: Undefined behaviour in
> > drivers/input/input.c:62:23 [  196.188958] signed integer overflow:
> > [  196.188964] -2147483647 - 104 cannot be represented in type 'int [2]'
> > [  196.188973] CPU: 7 PID: 4763 Comm: syz-executor Not tainted
> > 4.19.0-514.55.6.9.x86_64+ #7 [  196.188977] Hardware name: Bochs
> > Bochs, BIOS Bochs 01/01/2011 [  196.188979] Call Trace:
> > [  196.189001]  dump_stack+0x91/0xeb
> > [  196.189014]  ubsan_epilogue+0x9/0x7c [  196.189020]
> > handle_overflow+0x1d7/0x22c [  196.189028]  ?
> > __ubsan_handle_negate_overflow+0x18f/0x18f
> > [  196.189038]  ? __mutex_lock+0x213/0x13f0 [  196.189053]  ?
> > drop_futex_key_refs+0xa0/0xa0 [  196.189070]  ?
> > __might_fault+0xef/0x1b0 [  196.189096]
> > input_handle_event+0xe1b/0x1290 [  196.189108]
> > input_inject_event+0x1d7/0x27e [  196.189119]
> evdev_write+0x2cf/0x3f0
> > [  196.189129]  ? evdev_pass_values+0xd40/0xd40 [  196.189157]  ?
> > mark_held_locks+0x160/0x160 [  196.189171]  ?
> __vfs_write+0xe0/0x6c0 [
> > 196.189175]  ? evdev_pass_values+0xd40/0xd40 [  196.189179]
> > __vfs_write+0xe0/0x6c0 [  196.189186]  ? kernel_read+0x130/0x130 [
> > 196.189204]  ? _cond_resched+0x15/0x30 [  196.189214]  ?
> > __inode_security_revalidate+0xb8/0xe0
> > [  196.189222]  ? selinux_file_permission+0x354/0x430
> > [  196.189233]  vfs_write+0x160/0x440
> > [  196.189242]  ksys_write+0xc1/0x190
> > [  196.189248]  ? __ia32_sys_read+0xb0/0xb0 [  196.189259]  ?
> > trace_hardirqs_on_thunk+0x1a/0x1c [  196.189267]  ?
> > do_syscall_64+0x22/0x4a0 [  196.189276]  do_syscall_64+0xa5/0x4a0 [
> > 196.189287]  entry_SYSCALL_64_after_hwframe+0x49/0xbe
> > [  196.189293] RIP: 0033:0x44e7c9
> > [  196.189299] Code: fc ff 48 81 c4 80 00 00 00 e9 f1 fe ff ff 0f 1f
> > 00
> >
> > the syzkaller reproduce script(but can't reproduce it every time):
> >
> > r0 = syz_open_dev$evdev(&(0x7f0000000100)='/dev/input/event#\x00',
> > 0x2,
> > 0x1)
> > write$binfmt_elf64(r0, &(0x7f0000000240)={{0x7f, 0x45, 0x4c, 0x46,
> > 0x40, 0x2, 0x2, 0xffffffff, 0xffffffffffff374c, 0x3, 0x0, 0x80000001,
> > 0x103, 0x40, 0x22e, 0x26, 0x1, 0x38, 0x2, 0xa23, 0x1, 0x2},
> > [{0x6474e557, 0x5, 0x6, 0x2, 0x9, 0x9, 0x6c3, 0x1ff}], "", [[], [],
> > [], []]}, 0x478) ioctl$EVIOCGSW(0xffffffffffffffff, 0x8040451b,
> > &(0x7f0000000040)=""/7)
> > syz_open_dev$evdev(&(0x7f0000000100)='/dev/input/event#\x00', 0x2,
> > 0x1)
> > r1 = syz_open_dev$evdev(&(0x7f0000000100)='/dev/input/event#\x00',
> > 0x2,
> > 0x1)
> > openat$smack_task_current(0xffffffffffffff9c,
> > &(0x7f0000000040)='/proc/self/attr/current\x00', 0x2, 0x0)
> > ioctl$EVIOCSABS0(r1, 0x401845c0, &(0x7f0000000000)={0x4, 0x10000,
> 0x4,
> > 0xd1, 0x81, 0x3})
> > eventfd(0x1ff)
> > syz_open_dev$evdev(&(0x7f0000000100)='/dev/input/event#\x00', 0x2,
> > 0x200)
> > syz_open_dev$evdev(&(0x7f0000000100)='/dev/input/event#\x00', 0x2,
> > 0x1) syz_open_dev$evdev(&(0x7f0000000100)='/dev/input/event#\x00',
> > 0x2, 0x1)
> > syz_open_dev$evdev(&(0x7f0000000100)='/dev/input/event#\x00', 0x2,
> > 0x1) syz_open_dev$evdev(&(0x7f0000000100)='/dev/input/event#\x00',
> > 0x2, 0x1)
> >
> > Typecast int to long to fix the issue.
> 
> Does this fix 32-bit platforms where long equals int? BTW, I'd prefer if we did
> not have to do 64 bit arithmetic on 32 bit arches here, if possible.
Hi Dmitry,
Thanks for your review,  we can typecast it to long long?  but if do not 64 bit arithmetic on 32-bit platforms, how about this?
We can not care about the overflow of addition/subtraction, but we should care about the return value.?

diff --git a/drivers/input/input.c b/drivers/input/input.c
index 3304aaa..954244f 100644
--- a/drivers/input/input.c
+++ b/drivers/input/input.c
@@ -59,14 +59,17 @@ static inline int is_event_supported(unsigned int code,
 static int input_defuzz_abs_event(int value, int old_val, int fuzz)
 {
        if (fuzz) {
-               if (value > old_val - fuzz / 2 && value < old_val + fuzz / 2)
+               if (value > (long)((unsigned long)old_val - (unsigned long)fuzz / 2) &&
+                               value < (long)((unsigned long)old_val + (unsigned long)fuzz / 2))
                        return old_val;

-               if (value > old_val - fuzz && value < old_val + fuzz)
-                       return (old_val * 3 + value) / 4;
+               if (value > (long)((unsigned long)old_val - (unsigned long)fuzz) &&
+                               value < (long)((unsigned long)old_val + (unsigned long)fuzz))
+                       return ((long long)old_val * 3 + value) / 4;

-               if (value > old_val - fuzz * 2 && value < old_val + fuzz * 2)
-                       return (old_val + value) / 2;
+               if (value > (long)((unsigned long)old_val - (unsigned long)fuzz * 2) &&
+                               value < (long)((unsigned long)old_val + (unsigned long)fuzz * 2))
+                       return ((long long)old_val + value) / 2;
        }

        return value;


> Thanks.
> 
> --
> Dmitry