2022-11-10 03:51:36

by Zhen Lei

Subject: [PATCH] fs: Clear a UBSAN shift-out-of-bounds warning

UBSAN: shift-out-of-bounds in fs/locks.c:2572:16
left shift of 1 by 63 places cannot be represented in type 'long long int'

Switch the calculation method to ((quarter - 1) * 2 + 1) can help us
eliminate this false positive.

On the other hand, the old implementation has problems with char and
short types, although not currently involved.
printf("%d: %x\n", sizeof(char), INT_LIMIT(char));
printf("%d: %x\n", sizeof(short), INT_LIMIT(short));
1: ffffff7f
2: ffff7fff

Signed-off-by: Zhen Lei <[email protected]>
---
include/linux/fs.h | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/include/linux/fs.h b/include/linux/fs.h
index e654435f16512c1..88d42e2daed9f6c 100644
--- a/include/linux/fs.h
+++ b/include/linux/fs.h
@@ -1131,7 +1131,7 @@ struct file_lock_context {

/* The following constant reflects the upper bound of the file/locking space */
#ifndef OFFSET_MAX
-#define INT_LIMIT(x) (~((x)1 << (sizeof(x)*8 - 1)))
+#define INT_LIMIT(x) ((((x)1 << (sizeof(x) * 8 - 2)) - 1) * 2 + 1)
#define OFFSET_MAX INT_LIMIT(loff_t)
#define OFFT_OFFSET_MAX INT_LIMIT(off_t)
#endif
--
2.25.1

2022-11-19 20:04:46

by Eric Biggers

Subject: Re: [PATCH] fs: Clear a UBSAN shift-out-of-bounds warning

On Thu, Nov 10, 2022 at 11:10:24AM +0800, Zhen Lei wrote:
> UBSAN: shift-out-of-bounds in fs/locks.c:2572:16
> left shift of 1 by 63 places cannot be represented in type 'long long int'
>
> Switch the calculation method to ((quarter - 1) * 2 + 1) can help us
> eliminate this false positive.
>
> On the other hand, the old implementation has problems with char and
> short types, although not currently involved.
> printf("%d: %x\n", sizeof(char), INT_LIMIT(char));
> printf("%d: %x\n", sizeof(short), INT_LIMIT(short));
> 1: ffffff7f
> 2: ffff7fff
>
> Signed-off-by: Zhen Lei <[email protected]>
> ---
> include/linux/fs.h | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/include/linux/fs.h b/include/linux/fs.h
> index e654435f16512c1..88d42e2daed9f6c 100644
> --- a/include/linux/fs.h
> +++ b/include/linux/fs.h
> @@ -1131,7 +1131,7 @@ struct file_lock_context {
>
> /* The following constant reflects the upper bound of the file/locking space */
> #ifndef OFFSET_MAX
> -#define INT_LIMIT(x) (~((x)1 << (sizeof(x)*8 - 1)))
> +#define INT_LIMIT(x) ((((x)1 << (sizeof(x) * 8 - 2)) - 1) * 2 + 1)
> #define OFFSET_MAX INT_LIMIT(loff_t)
> #define OFFT_OFFSET_MAX INT_LIMIT(off_t)
> #endif

This problem has already been solved by type_max() in include/linux/overflow.h.
How about removing INT_LIMIT() and using type_max() instead?

- Eric

2022-11-21 02:19:08

by Zhen Lei

Subject: Re: [PATCH] fs: Clear a UBSAN shift-out-of-bounds warning

On 2022/11/20 3:39, Eric Biggers wrote:
> On Thu, Nov 10, 2022 at 11:10:24AM +0800, Zhen Lei wrote:
>> UBSAN: shift-out-of-bounds in fs/locks.c:2572:16
>> left shift of 1 by 63 places cannot be represented in type 'long long int'
>>
>> Switch the calculation method to ((quarter - 1) * 2 + 1) can help us
>> eliminate this false positive.
>>
>> On the other hand, the old implementation has problems with char and
>> short types, although not currently involved.
>> printf("%d: %x\n", sizeof(char), INT_LIMIT(char));
>> printf("%d: %x\n", sizeof(short), INT_LIMIT(short));
>> 1: ffffff7f
>> 2: ffff7fff
>>
>> Signed-off-by: Zhen Lei <[email protected]>
>> ---
>> include/linux/fs.h | 2 +-
>> 1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/include/linux/fs.h b/include/linux/fs.h
>> index e654435f16512c1..88d42e2daed9f6c 100644
>> --- a/include/linux/fs.h
>> +++ b/include/linux/fs.h
>> @@ -1131,7 +1131,7 @@ struct file_lock_context {
>>
>> /* The following constant reflects the upper bound of the file/locking space */
>> #ifndef OFFSET_MAX
>> -#define INT_LIMIT(x) (~((x)1 << (sizeof(x)*8 - 1)))
>> +#define INT_LIMIT(x) ((((x)1 << (sizeof(x) * 8 - 2)) - 1) * 2 + 1)
>> #define OFFSET_MAX INT_LIMIT(loff_t)
>> #define OFFT_OFFSET_MAX INT_LIMIT(off_t)
>> #endif
>
> This problem has already been solved by type_max() in include/linux/overflow.h.
> How about removing INT_LIMIT() and using type_max() instead?

It's better to use type_max(). INT_LIMIT() is currently valid only for signed type.
Besides, my method is actually the same as type_max(). Using type_max() does not
result in code duplication.

>
> - Eric
> .
>

--
Regards,
Zhen Lei