[RFC PATCH 1/2] log2: handle LARGE input to __roundup_pow_of_two()

[RFC PATCH 1/2] log2: handle LARGE input to __roundup_pow_of_two()

[Randy Dunlap]

UBSAN detected a 64-bit shift in log2.h:__roundup_pow_of_two():
  UBSAN: shift-out-of-bounds in ./include/linux/log2.h:57:13
  shift exponent 64 is too large for 64-bit type 'long unsigned int'

This is during a call from mm/readahead.c:ondemand_readahead(),
get_init_ra_size(), where the 'size' parameter must have been
extremely large (or "negative").

fls() can legitimately return 32 or 64 when the MSbit is set in
a 32-bit or 64-bit unsigned long. For these values, doing
"1UL << shiftcout" is invalid or undefined, so catch when this
happens.

When the MSbit is 32 or 64, we cannot roundup to the next power of 2,
so just return n (the input value), which is >= 0x8000...0000 and
probably not a power of 2 (unless it is exactly 0x8000...0000).

Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Toralf Förster <toralf.foerster@gmx.de>
Cc: linux-mm@kvack.org
---
 include/linux/log2.h |   12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)

--- linux-5.10.1.orig/include/linux/log2.h
+++ linux-5.10.1/include/linux/log2.h
@@ -54,7 +54,17 @@ bool is_power_of_2(unsigned long n)
 static inline __attribute__((const))
 unsigned long __roundup_pow_of_two(unsigned long n)
 {
-	return 1UL << fls_long(n - 1);
+	unsigned int lastset = fls_long(n - 1); /* this can be 64 or 32 */
+
+	/*
+	 * for high bit set (64 or 32), we can neither round up nor
+	 * make it a power or 2
+	 */
+	if ((sizeof(n) == 4 && lastset == 32) ||
+	    (sizeof(n) == 8 && lastset == 64))
+		return n;
+
+	return 1UL << lastset;
 }
 
 /**

Re: [RFC PATCH 1/2] log2: handle LARGE input to __roundup_pow_of_two()

[Matthew Wilcox]

On Sun, Dec 20, 2020 at 01:10:37PM -0800, Randy Dunlap wrote:
> UBSAN detected a 64-bit shift in log2.h:__roundup_pow_of_two():
>   UBSAN: shift-out-of-bounds in ./include/linux/log2.h:57:13
>   shift exponent 64 is too large for 64-bit type 'long unsigned int'
> 
> This is during a call from mm/readahead.c:ondemand_readahead(),
> get_init_ra_size(), where the 'size' parameter must have been
> extremely large (or "negative").

Actually, I think it was zero, which is the real bug that should be fixed.

Re: [RFC PATCH 1/2] log2: handle LARGE input to __roundup_pow_of_two()

[Randy Dunlap]

On 12/20/20 1:23 PM, Matthew Wilcox wrote:
> On Sun, Dec 20, 2020 at 01:10:37PM -0800, Randy Dunlap wrote:
>> UBSAN detected a 64-bit shift in log2.h:__roundup_pow_of_two():
>>   UBSAN: shift-out-of-bounds in ./include/linux/log2.h:57:13
>>   shift exponent 64 is too large for 64-bit type 'long unsigned int'
>>
>> This is during a call from mm/readahead.c:ondemand_readahead(),
>> get_init_ra_size(), where the 'size' parameter must have been
>> extremely large (or "negative").
> 
> Actually, I think it was zero, which is the real bug that should be fixed.
> 

Hm, OK, that would make more sense than some Huge value (other than -1).

Do you mean something like this?

---
---
 mm/readahead.c |    6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

--- linux-5.10.1.orig/mm/readahead.c
+++ linux-5.10.1/mm/readahead.c
@@ -310,7 +310,11 @@ void force_page_cache_ra(struct readahea
  */
 static unsigned long get_init_ra_size(unsigned long size, unsigned long max)
 {
-	unsigned long newsize = roundup_pow_of_two(size);
+	unsigned long newsize;
+
+	if (!size)
+		size = 32;
+	newsize = roundup_pow_of_two(size);
 
 	if (newsize <= max / 32)
 		newsize = newsize * 4;


Thanks.