[PATCH] loop: avoid out-of-range warning

From: Arnd Bergmann <arnd@arndb.de>

clang warns that the sanity check for page size always succeeds
when building with 64KB pages:

drivers/block/loop.c:282:27: error: result of comparison of constant 65536 with expression of type 'unsigned short' is always false [-Werror,-Wtautological-constant-out-of-range-compare]
        if (bsize < 512 || bsize > PAGE_SIZE || !is_power_of_2(bsize))
                           ~~~~~ ^ ~~~~~~~~~

There is nothing wrong here, so just shut up the check by changing
the type of the bsize argument.

Fixes: 3448914e8cc5 ("loop: Add LOOP_CONFIGURE ioctl")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
---
 drivers/block/loop.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/block/loop.c b/drivers/block/loop.c
index 7bf4686af774..51315a93b399 100644
--- a/drivers/block/loop.c
+++ b/drivers/block/loop.c

@@ -277,7 +277,7 @@ static void __loop_update_dio(struct loop_device *lo, bool dio)
  * @bsize: size to validate
  */
 static int
-loop_validate_block_size(unsigned short bsize)
+loop_validate_block_size(unsigned int bsize)
 {
 	if (bsize < 512 || bsize > PAGE_SIZE || !is_power_of_2(bsize))
 		return -EINVAL;

-- 
2.29.2

Thanks!

On Mon, Sep 27, 2021 at 11:43 AM Arnd Bergmann [off-list ref] wrote:

From: Arnd Bergmann <arnd@arndb.de>

clang warns that the sanity check for page size always succeeds
when building with 64KB pages:

drivers/block/loop.c:282:27: error: result of comparison of constant 65536 with expression of type 'unsigned short' is always false [-Werror,-Wtautological-constant-out-of-range-compare]
        if (bsize < 512 || bsize > PAGE_SIZE || !is_power_of_2(bsize))
                           ~~~~~ ^ ~~~~~~~~~

There is nothing wrong here, so just shut up the check by changing
the type of the bsize argument.

Fixes: 3448914e8cc5 ("loop: Add LOOP_CONFIGURE ioctl")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>

Reviewed-by: Martijkn Coenen <redacted>

Well, you found a bug here.

lo_simple_ioctl(LOOP_SET_BLOCK_SIZE) passes "unsigned long arg" to
loop_validate_block_size() via loop_set_block_size(), and uses only
lower 32bits. "(lo->lo_queue->limits.logical_block_size == arg)" is
failing to "return 0;" if lower 32bits are the same but higher 32bits
are not 0. Shouldn't we also check that higher 32bits are 0, by either
making loop_validate_block_size() to accept "unsigned long" or making
loop_set_block_size() return -EINVAL if larger than UINT_MAX ?

There is nothing wrong here, so just shut up the check by changing
the type of the bsize argument.

I think there is something wrong here.
Since "unsigned short bsize" checked only lower 16bits, a fuzzer can
pass e.g. 0xDEAD0200 (which is not a is_power_of_2() 32bits value) to

	blk_queue_logical_block_size(lo->lo_queue, arg);
	blk_queue_physical_block_size(lo->lo_queue, arg);
	blk_queue_io_min(lo->lo_queue, arg);

and confuse the system?