Skip to contentSkip to conversation

hfsplus: fix uninit-value in hfsplus_strcasecmp

PATCH In review Deepanshu Kartikey 2 participants 2 messages Jan 21

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 hfsplus: fix uninit-value in hfsplus_strcasecmp

Syzbot reported a KMSAN uninit-value issue in hfsplus_strcasecmp() during filesystem mount operations. The root cause is that hfsplus_find_cat() declares a local hfsplus_cat_entry variable without initialization before passing it to hfs_brec_read().

When the filesystem image is corrupted or malformed (as syzbot fuzzes), hfs_brec_read() may read less data than sizeof(hfsplus_cat_entry). In such cases, the tmp.thread.nodeName.unicode array may only be partially filled, leaving remaining bytes uninitialized.

hfsplus_cat_build_key_uni() then copies from this array based on nodeName.length. If the on-disk length field is corrupted or the array wasn't fully written by hfs_brec_read(), uninitialized stack data gets copied into the search key. When hfsplus_strcasecmp() subsequently reads these uninitialized bytes and uses them in case_fold() as an array index into hfsplus_case_fold_table, KMSAN detects the use of uninitialized values.

Fix this by initializing tmp to zero, ensuring that even with corrupted filesystem images, no uninitialized data is propagated.

---
Changes in v2:
- Use structure initialization (= {0}) instead of memset() as suggested by Viacheslav Dubeyko
- Improved commit message to clarify how uninitialized data is used

---
 fs/hfsplus/catalog.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
fs/hfsplus/catalog.c +1 -1
index 02c1eee4a4b8..56a53d2d437e 100644
@@ -194,7 +194,7 @@ static int hfsplus_fill_cat_thread(struct super_block *sb,
 int hfsplus_find_cat(struct super_block *sb, u32 cnid,
 		     struct hfs_find_data *fd)
 {
-	hfsplus_cat_entry tmp;
+	hfsplus_cat_entry tmp = {0};
 	int err;
 	u16 type;
 
Signature · 1 line
2.43.0
#1 lore raw
Quoted 9 lines · Deepanshu Kartikey

Syzbot reported a KMSAN uninit-value issue in hfsplus_strcasecmp() during filesystem mount operations. The root cause is that hfsplus_find_cat() declares a local hfsplus_cat_entry variable without initialization before passing it to hfs_brec_read().

When the filesystem image is corrupted or malformed (as syzbot fuzzes), hfs_brec_read() may read less data than sizeof(hfsplus_cat_entry). In such cases, the tmp.thread.nodeName.unicode array may only be partially filled, leaving remaining bytes uninitialized.

I still don't understand this point. Please, see my questions in the thread for v1 of the patch. Could you please address my questions/worries?

Thanks,
Slava.

Quoted 39 lines

hfsplus_cat_build_key_uni() then copies from this array based on nodeName.length. If the on-disk length field is corrupted or the array wasn't fully written by hfs_brec_read(), uninitialized stack data gets copied into the search key. When hfsplus_strcasecmp() subsequently reads these uninitialized bytes and uses them in case_fold() as an array index into hfsplus_case_fold_table, KMSAN detects the use of uninitialized values.

Fix this by initializing tmp to zero, ensuring that even with corrupted filesystem images, no uninitialized data is propagated.

---
Changes in v2:
- Use structure initialization (= {0}) instead of memset() as suggested by Viacheslav Dubeyko
- Improved commit message to clarify how uninitialized data is used

---
 fs/hfsplus/catalog.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
fs/hfsplus/catalog.c +1 -1
index 02c1eee4a4b8..56a53d2d437e 100644
@@ -194,7 +194,7 @@ static int hfsplus_fill_cat_thread(struct super_block *sb,
 int hfsplus_find_cat(struct super_block *sb, u32 cnid,
 		     struct hfs_find_data *fd)
 {
-	hfsplus_cat_entry tmp;
+	hfsplus_cat_entry tmp = {0};
 	int err;
 	u16 type;