udf: potential integer overflow

PATCH In review Dan Carpenter 2 participants 2 messages Mar 15, 2010

Subsystems (2)

Kernel janitors UDF filesystem

lore

Share & tools

lore

Dan Carpenter

author

2010-03-15 08:21 UTC #0 lore raw

PATCH 1/1 udf: potential integer overflow

bloc->logicalBlockNum is unsigned so it's never less than zero.

When I saw that, it made me worry that "bloc->logicalBlockNum + count" could overflow. That's why I changed the check for less than zero to an overflow check. (The test works because "count" is also unsigned.)

Signed-off-by Dan Carpenter <[email protected]>

---
GCC 4.1 apparently optimizes overflow checks like this away, but it should work for other versions of gcc. I tested with GCC 4.3.
http://www.fefe.de/intof.html

fs/udf/balloc.c +4 -6

index 19626e2..9a9378b 100644
@@ -125,9 +125,8 @@ static void udf_bitmap_free_blocks(struct super_block *sb,
 
 	mutex_lock(&sbi->s_alloc_mutex);
 	partmap = &sbi->s_partmaps[bloc->partitionReferenceNum];
-	if (bloc->logicalBlockNum < 0 ||
-	    (bloc->logicalBlockNum + count) >
-		partmap->s_partition_len) {
+	if (bloc->logicalBlockNum + count < count ||
+	    (bloc->logicalBlockNum + count) > partmap->s_partition_len) {
 		udf_debug("%d < %d || %d + %d > %d\n",
 			  bloc->logicalBlockNum, 0, bloc->logicalBlockNum,
 			  count, partmap->s_partition_len);

@@ -393,9 +392,8 @@ static void udf_table_free_blocks(struct super_block *sb,
 
 	mutex_lock(&sbi->s_alloc_mutex);
 	partmap = &sbi->s_partmaps[bloc->partitionReferenceNum];
-	if (bloc->logicalBlockNum < 0 ||
-	    (bloc->logicalBlockNum + count) >
-		partmap->s_partition_len) {
+	if (bloc->logicalBlockNum + count < count ||
+	    (bloc->logicalBlockNum + count) > partmap->s_partition_len) {
 		udf_debug("%d < %d || %d + %d > %d\n",
 			  bloc->logicalBlockNum, 0, bloc->logicalBlockNum, count,
 			  partmap->s_partition_len);

Jan Kara

2010-03-15 12:08 UTC #1 lore raw

Quoted 8 lines · Dan Carpenter

bloc->logicalBlockNum is unsigned so it's never less than zero.
When I saw that, it made me worry that "bloc->logicalBlockNum + count" could overflow. That's why I changed the check for less than zero to an overflow check. (The test works because "count" is also unsigned.)
Signed-off-by Dan Carpenter <[email protected]>

Thanks. Merged.

Quoted 4 lines

---
GCC 4.1 apparently optimizes overflow checks like this away, but it should work for other versions of gcc. I tested with GCC 4.3.
http://www.fefe.de/intof.html

It should only optimize them out for signed types (moreover kernel has
this optimization turned off so it's a non-issue for us anyway).

Honza

Signature · 2 lines

Jan Kara <[email protected]>
SUSE Labs, CR