Skip to contentSkip to conversation

fs: btrfs: prevent unintentional int overflow

PATCH In review Cengiz Can 2 participants 5 messages Jan 3, 2020 – Jan 7, 2020 Btrfs file system

lore
Share & tools
lore

Cengiz Can

author
#0 lore raw

PATCH 1/1 fs: btrfs: prevent unintentional int overflow

Coverity scan for 5.5.0-rc4 found a possible integer overflow in tree-checker.c line 364.

prev_csum_end = (prev_item_size / csumsize) * sectorsize;

Quoting from scan results:

```
CID 1456959 Unintentional integer overflow

Unintentional integer overflow (OVERFLOW_BEFORE_WIDEN) overflow_before_widen:
Potentially overflowing expression `prev_item_size / csumsize * sectorsize`
with type unsigned int (32 bits, unsigned) is evaluated using 32-bit
arithmetic, and then used in a context that expects an expression of type u64.
(64 bits, unsigned).
```

Added a cast to u64 on the left hand side of the expression.

Compiles fine on x86_64_defconfig with all btrfs config flags enabled.

---
 fs/btrfs/tree-checker.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
fs/btrfs/tree-checker.c +1 -1
index 97f3520b8d98..9f58f07be779 100644
@@ -361,7 +361,7 @@ static int check_csum_item(struct extent_buffer *leaf, struct btrfs_key *key,
 		u32 prev_item_size;
 
 		prev_item_size = btrfs_item_size_nr(leaf, slot - 1);
-		prev_csum_end = (prev_item_size / csumsize) * sectorsize;
+		prev_csum_end = (u64) (prev_item_size / csumsize) * sectorsize;
 		prev_csum_end += prev_key->offset;
 		if (prev_csum_end > key->offset) {
 			generic_err(leaf, slot - 1,
Signature · 1 line
2.20.1
#1 lore raw
Quoted 31 lines · Cengiz Can

Coverity scan for 5.5.0-rc4 found a possible integer overflow in tree-checker.c line 364.

prev_csum_end = (prev_item_size / csumsize) * sectorsize;

Quoting from scan results:

```
CID 1456959 Unintentional integer overflow

Unintentional integer overflow (OVERFLOW_BEFORE_WIDEN) overflow_before_widen:
Potentially overflowing expression `prev_item_size / csumsize * sectorsize`
with type unsigned int (32 bits, unsigned) is evaluated using 32-bit
arithmetic, and then used in a context that expects an expression of type u64.
(64 bits, unsigned).
```

Added a cast to u64 on the left hand side of the expression.

Compiles fine on x86_64_defconfig with all btrfs config flags enabled.

---
 fs/btrfs/tree-checker.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
fs/btrfs/tree-checker.c +0 -0
index 97f3520b8d98..9f58f07be779 100644

fs/btrfs/tree-checker.c static int check_csum_item(struct extent_buffer *leaf, struct btrfs_key *key, · line 364

Quoted 5 lines
fs/btrfs/tree-checker.c +1 -1
@@ -361,7 +361,7 @@ static int check_csum_item(struct extent_buffer *leaf, struct btrfs_key *key,
 		u32 prev_item_size;
 
 		prev_item_size = btrfs_item_size_nr(leaf, slot - 1);
-		prev_csum_end = (prev_item_size / csumsize) * sectorsize;
+		prev_csum_end = (u64) (prev_item_size / csumsize) * sectorsize;

The overflow can't happen in practice. Taking generous maximum value for an item and sectorsize (64K) and doing the math will reach nowhere the overflow limit for 32bit type:

64K / 4 * 64K = 1G

I'm not sure if this is worth adding the cast, or mark the coverity issue as not important.

author
#2 lore raw
Quoted 5 lines · On January 6, 2020 18:53:40 David Sterba <[email protected]> wrote:

The overflow can't happen in practice. Taking generous maximum value for an item and sectorsize (64K) and doing the math will reach nowhere the overflow limit for 32bit type:

64K / 4 * 64K = 1G

I didn't know that. Thanks for sharing.

Quoted 2 lines

I'm not sure if this is worth adding the cast, or mark the coverity issue as not important.

If the cast is adding any overhead (which I don't think it does) I would kindly request adding it for the sake of completeness.

For example: if some newbie like me tries adding something, they would be warned that we should consider possible overflows and/or at least be careful with expressions that contain different sized integers.

If this sounds unnecessary, I will help with marking the Coverity issue as unimportant.

Thank you!

#3 lore raw
Quoted 8 lines · Cengiz Can
Quoted 5 lines · On January 6, 2020 18:53:40 David Sterba <[email protected]> wrote:

The overflow can't happen in practice. Taking generous maximum value for an item and sectorsize (64K) and doing the math will reach nowhere the overflow limit for 32bit type:

64K / 4 * 64K = 1G

Quoted 7 lines

I didn't know that. Thanks for sharing.

Quoted 2 lines

I'm not sure if this is worth adding the cast, or mark the coverity issue as not important.

If the cast is adding any overhead (which I don't think it does) I would kindly request adding it for the sake of completeness.

It's not a runtime overhead but typecasts should not be needed in general and when there's one it should be there for a reason. Sometimes it's apparent and does not need a comment to explain why but otherwise I see it as "overhead" when reading the code. Lots of calculations done in btrfs fit perfectly to 32bit, eg. the b-tree node or page-related ones. Where it matters is eg. conversion from/to bio::bi_sector to/from btrfs logical addresses that are u64, where the interface type is unsigned long and not a fixed width.

The size constraints of the variables used in the reported expression are known to developers so I tend to think the typecast is not necessary. Maybe the static checker tool could be improved to know the invariants, that are eg. verified in fs/btrfs/disk-io.c:validate_super.

author
#4 lore raw

Hello David!

Quoted 8 lines · David Sterba

It's not a runtime overhead but typecasts should not be needed in general and when there's one it should be there for a reason. Sometimes it's apparent and does not need a comment to explain why but otherwise I see it as "overhead" when reading the code. Lots of calculations done in btrfs fit perfectly to 32bit, eg. the b-tree node or page-related ones. Where it matters is eg. conversion from/to bio::bi_sector to/from btrfs logical addresses that are u64, where the interface type is unsigned long and not a fixed width.

Thanks for sharing that. As I said, I'm relatively new to btrfs internals.

Quoted 3 lines

The size constraints of the variables used in the reported expression are known to developers so I tend to think the typecast is not necessary.

Agreed.

Quoted 2 lines

Maybe the static checker tool could be improved to know the invariants, that are eg. verified in fs/btrfs/disk-io.c:validate_super.

That's something that I will do some research on.

We can ignore this patch.

Thank you!

Signature · 2 lines
Cengiz Can
@cengiz_io