[PATCH] fs/buffer.c: grow_buffers: fix the uncorrect check

[PATCH] fs/buffer.c: grow_buffers: fix the uncorrect check

[mnipxh]

When pgoff_t index is 32bit, sector_t block is 64bit, need check if block number is too big.
If block is bigger than (4Gb * PAGE_SIZE), index becomes a wrong value.
Commit e5657933863f43cc6bb76a54d659303dafaa9e58 wants to do this. But it gives an uncorrect check.
I think block != index << sizebits is correct. And it can detect such issue above.

Signed-off-by: xinhui.pan <xinhuix.pan@intel.com>
---
 fs/buffer.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/fs/buffer.c b/fs/buffer.c
index 9ddb9fc..1a674a6 100644
--- a/fs/buffer.c
+++ b/fs/buffer.c
@@ -1081,7 +1081,7 @@ grow_buffers(struct block_device *bdev, sector_t block, int size)
 	 * Check for a block which wants to lie outside our maximum possible
 	 * pagecache index.  (this comparison is done using sector_t types).
 	 */
-	if (unlikely(index != block >> sizebits)) {
+	if (unlikely(block != index << sizebits)) {
 		char b[BDEVNAME_SIZE];
 
 		printk(KERN_ERR "%s: requested out-of-range block %llu for "
-- 
1.9.1

Re: [PATCH] fs/buffer.c: grow_buffers: fix the uncorrect check

[Linus Torvalds]

On Sat, May 31, 2014 at 11:12 AM, mnipxh <mnipxh@gmail.com> wrote:
> When pgoff_t index is 32bit, sector_t block is 64bit, need check if block number is too big.
> If block is bigger than (4Gb * PAGE_SIZE), index becomes a wrong value.
> Commit e5657933863f43cc6bb76a54d659303dafaa9e58 wants to do this. But it gives an uncorrect check.
> I think block != index << sizebits is correct. And it can detect such issue above.

I don't understand why you think the current check is incorrect.

It is testing that the calculation hasn't overflowed. It is correct.
It *should* use the same calculation, to (a) make it more obvious that
it's double-checking the earlier calculation in a different type, and
(b) to make it easier for the compiler to optimize it away if sector_t
and pgoff_t are the same size.

So the current code is correct, afaik. Note that "index" is "pgoff_t",
but "block >> sizebits" is "sector_t".

               Linus