[PATCH] ntfs3: fix uninit memory after failed mi_read in mi_format_new

[PATCH] ntfs3: fix uninit memory after failed mi_read in mi_format_new

[Raphael Pinsonneault-Thibeault]

attr_insert_range() called from ntfs_fallocate() has 2 different
code paths that trigger mi_read() (which calls ntfs_read_bh).
If the first mi_read() -> ntfs_read_bh() fails with an IO error, it
leaves an uninitialized buffer in the buffer cache.
The second mi_read() -> ntfs_read_bh() then uses that buffer,
where we get KMSAN warning "uninit-value in ntfs_read_bh".

The fix is to check if mi_read failed in mi_format_new.

Reported-by: syzbot+7a2ba6b7b66340cff225@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=7a2ba6b7b66340cff225
Tested-by: syzbot+7a2ba6b7b66340cff225@syzkaller.appspotmail.com
Fixes: 4342306f0f0d5 ("fs/ntfs3: Add file operations and implementation")
Signed-off-by: Raphael Pinsonneault-Thibeault <rpthibeault@gmail.com>
---
 fs/ntfs3/record.c | 19 ++++++++++++-------
 1 file changed, 12 insertions(+), 7 deletions(-)

diff --git a/fs/ntfs3/record.c b/fs/ntfs3/record.c
index 714c7ecedca8..98d2e5517077 100644
--- a/fs/ntfs3/record.c
+++ b/fs/ntfs3/record.c
@@ -431,13 +431,18 @@ int mi_format_new(struct mft_inode *mi, struct ntfs_sb_info *sbi, CLST rno,
 		seq = rno;
 	} else if (rno >= sbi->mft.used) {
 		;
-	} else if (mi_read(mi, is_mft)) {
-		;
-	} else if (rec->rhdr.sign == NTFS_FILE_SIGNATURE) {
-		/* Record is reused. Update its sequence number. */
-		seq = le16_to_cpu(rec->seq) + 1;
-		if (!seq)
-			seq = 1;
+	} else {
+		err = mi_read(mi, is_mft);
+		if (err) {
+			return err;
+		}
+
+		if (rec->rhdr.sign == NTFS_FILE_SIGNATURE) {
+			/* Record is reused. Update its sequence number. */
+			seq = le16_to_cpu(rec->seq) + 1;
+			if (!seq)
+				seq = 1;
+		}
 	}
 
 	memcpy(rec, sbi->new_rec, sbi->record_size);
-- 
2.43.0

Re: [PATCH] ntfs3: fix uninit memory after failed mi_read in mi_format_new

[David Hunter]

On 9/25/25 16:36, Raphael Pinsonneault-Thibeault wrote:
> attr_insert_range() called from ntfs_fallocate() has 2 different
> code paths that trigger mi_read() (which calls ntfs_read_bh).
> If the first mi_read() -> ntfs_read_bh() fails with an IO error, it
> leaves an uninitialized buffer in the buffer cache.
> The second mi_read() -> ntfs_read_bh() then uses that buffer,
> where we get KMSAN warning "uninit-value in ntfs_read_bh".
> 
> The fix is to check if mi_read failed in mi_format_new.
> 


Did you do any testing for this patch?

David Hunter

Re: [PATCH] ntfs3: fix uninit memory after failed mi_read in mi_format_new

[David Hunter]

On 9/25/25 16:36, Raphael Pinsonneault-Thibeault wrote:

> -	} else if (mi_read(mi, is_mft)) {
> -		;
> -	} else if (rec->rhdr.sign == NTFS_FILE_SIGNATURE) {
> -		/* Record is reused. Update its sequence number. */
> -		seq = le16_to_cpu(rec->seq) + 1;
> -		if (!seq)
> -			seq = 1;
> +	} else {
> +		err = mi_read(mi, is_mft);
> +		if (err) {
> +			return err;
> +		}
> +
> +		if (rec->rhdr.sign == NTFS_FILE_SIGNATURE) {
> +			/* Record is reused. Update its sequence number. */
> +			seq = le16_to_cpu(rec->seq) + 1;
> +			if (!seq)
> +				seq = 1;
> +		}

How was this tested?