Skip to contentSkip to conversation

btrfs: tree-checker: missing returns after data_ref alignment checks

RFC PATCH In review David Sterba 2 participants 3 messages Nov 16, 2020 – Nov 17, 2020 Btrfs file system

lore
Share & tools
lore
#0 lore raw

RFC PATCH 1/1 btrfs: tree-checker: missing returns after data_ref alignment checks

Hi Qu,

I've found more missing return satetments in tree-checker but I'm not sure if leaving them out was intentional. Both are for extent data_ref item alignment checks. It could be that alignment is not a hard problem, tough it could point to one, there's a check of the hash vs key->offset that would catch inconsistent data.

As there's only one statement after if, this looks like it was forgotten, but otherwise needs a comment why the returns are not there given that the rest of the file follows the same pattern "if / extent_err/ return -EUCLEAN".

---
 fs/btrfs/tree-checker.c | 2 ++
 1 file changed, 2 insertions(+)
fs/btrfs/tree-checker.c +2 -0
index 1b27242a9c0b..f3f666b343ef 100644
@@ -1424,6 +1424,7 @@ static int check_extent_data_ref(struct extent_buffer *leaf,
 	"invalid item size, have %u expect aligned to %zu for key type %u",
 			    btrfs_item_size_nr(leaf, slot),
 			    sizeof(*dref), key->type);
+		return -EUCLEAN;
 	}
 	if (!IS_ALIGNED(key->objectid, leaf->fs_info->sectorsize)) {
 		generic_err(leaf, slot,
@@ -1452,6 +1453,7 @@ static int check_extent_data_ref(struct extent_buffer *leaf,
 			extent_err(leaf, slot,
 	"invalid extent data backref offset, have %llu expect aligned to %u",
 				   offset, leaf->fs_info->sectorsize);
+			return -EUCLEAN;
 		}
 	}
 	return 0;
Signature · 1 line
2.25.0
#1 lore raw
Quoted 7 lines · David Sterba

Hi Qu,

I've found more missing return satetments in tree-checker but I'm not sure if leaving them out was intentional. Both are for extent data_ref item alignment checks. It could be that alignment is not a hard problem, tough it could point to one, there's a check of the hash vs key->offset that would catch inconsistent data.

All my bad.

The case is, alignment is the first sign of corruption. So if alignment is incorrect, there is no need to continue.

So yeah, I forgot more return lines.

Quoted 15 lines

As there's only one statement after if, this looks like it was forgotten, but otherwise needs a comment why the returns are not there given that the rest of the file follows the same pattern "if / extent_err/ return -EUCLEAN".

---
 fs/btrfs/tree-checker.c | 2 ++
 1 file changed, 2 insertions(+)
fs/btrfs/tree-checker.c +0 -0
index 1b27242a9c0b..f3f666b343ef 100644

fs/btrfs/tree-checker.c static int check_extent_data_ref(struct extent_buffer *leaf, · line 1427

Quoted 4 lines
fs/btrfs/tree-checker.c +1 -0
@@ -1424,6 +1424,7 @@ static int check_extent_data_ref(struct extent_buffer *leaf,
 	"invalid item size, have %u expect aligned to %zu for key type %u",
 			    btrfs_item_size_nr(leaf, slot),
 			    sizeof(*dref), key->type);
+		return -EUCLEAN;

For EXTENT_DATA_REF_KEY, it should only contains one or more btrfs_extent_data_ref, thus the alignment is a hard requirement.

Thus the return is needed.

fs/btrfs/tree-checker.c static int check_extent_data_ref(struct extent_buffer *leaf, · line 1456

Quoted 8 lines
fs/btrfs/tree-checker.c +1 -0
 	}
 	if (!IS_ALIGNED(key->objectid, leaf->fs_info->sectorsize)) {
 		generic_err(leaf, slot,
@@ -1452,6 +1453,7 @@ static int check_extent_data_ref(struct extent_buffer *leaf,
 			extent_err(leaf, slot,
 	"invalid extent data backref offset, have %llu expect aligned to %u",
 				   offset, leaf->fs_info->sectorsize);
+			return -EUCLEAN;

Basic sector size alignment requirement.
So yet another missing return.

Thanks,
Qu

Quoted 4 lines
fs/btrfs/tree-checker.c +0 -0
 		}
 	}
 	return 0;
#2 lore raw
Quoted 41 lines · Qu Wenruo
Quoted 7 lines · David Sterba

Hi Qu,

I've found more missing return satetments in tree-checker but I'm not sure if leaving them out was intentional. Both are for extent data_ref item alignment checks. It could be that alignment is not a hard problem, tough it could point to one, there's a check of the hash vs key->offset that would catch inconsistent data.

All my bad.

The case is, alignment is the first sign of corruption. So if alignment is incorrect, there is no need to continue.

So yeah, I forgot more return lines.

Quoted 15 lines

As there's only one statement after if, this looks like it was forgotten, but otherwise needs a comment why the returns are not there given that the rest of the file follows the same pattern "if / extent_err/ return -EUCLEAN".

---
 fs/btrfs/tree-checker.c | 2 ++
 1 file changed, 2 insertions(+)
fs/btrfs/tree-checker.c +0 -0
index 1b27242a9c0b..f3f666b343ef 100644

fs/btrfs/tree-checker.c static int check_extent_data_ref(struct extent_buffer *leaf, · line 1427

Quoted 4 lines
fs/btrfs/tree-checker.c +1 -0
@@ -1424,6 +1424,7 @@ static int check_extent_data_ref(struct extent_buffer *leaf,
 	"invalid item size, have %u expect aligned to %zu for key type %u",
 			    btrfs_item_size_nr(leaf, slot),
 			    sizeof(*dref), key->type);
+		return -EUCLEAN;

For EXTENT_DATA_REF_KEY, it should only contains one or more btrfs_extent_data_ref, thus the alignment is a hard requirement.

Thus the return is needed.

Quoted 11 lines

fs/btrfs/tree-checker.c static int check_extent_data_ref(struct extent_buffer *leaf, · line 1456

Quoted 8 lines
fs/btrfs/tree-checker.c +1 -0
 	}
 	if (!IS_ALIGNED(key->objectid, leaf->fs_info->sectorsize)) {
 		generic_err(leaf, slot,
@@ -1452,6 +1453,7 @@ static int check_extent_data_ref(struct extent_buffer *leaf,
 			extent_err(leaf, slot,
 	"invalid extent data backref offset, have %llu expect aligned to %u",
 				   offset, leaf->fs_info->sectorsize);
+			return -EUCLEAN;

Basic sector size alignment requirement.
So yet another missing return.

Thanks, I'll resend it as a proper patch.