linux-kernel - [PATCH v2] ntfs3: fix uninit memory after failed mi_read in mi_format

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [day] [month] [year] [list]

Message-ID: <20251012201635.378419-2-rpthibeault@gmail.com>
Date: Sun, 12 Oct 2025 16:16:34 -0400
From: Raphael Pinsonneault-Thibeault <rpthibeault@...il.com>
To: almaz.alexandrovich@...agon-software.com
Cc: Raphael Pinsonneault-Thibeault <rpthibeault@...il.com>,
	ntfs3@...ts.linux.dev,
	linux-kernel@...r.kernel.org,
	syzbot+7a2ba6b7b66340cff225@...kaller.appspotmail.com,
	linux-kernel-mentees@...ts.linux.dev,
	skhan@...uxfoundation.org,
	david.hunter.linux@...il.com,
	khalid@...nel.org
Subject: [PATCH v2] ntfs3: fix uninit memory after failed mi_read in mi_format_new

Fix a KMSAN un-init bug found by syzkaller.

ntfs_get_bh() expects a buffer from sb_getblk(), that buffer may not be
uptodate. We do not bring the buffer uptodate before setting it as
uptodate. If the buffer were to not be uptodate, it could mean adding a
buffer with un-init data to the mi record. Attempting to load that record
will trigger KMSAN.

Avoid this by setting the buffer as uptodate, if it’s not already, by
overwriting it.

Reported-by: syzbot+7a2ba6b7b66340cff225@...kaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=7a2ba6b7b66340cff225
Tested-by: syzbot+7a2ba6b7b66340cff225@...kaller.appspotmail.com
Fixes: 4342306f0f0d5 ("fs/ntfs3: Add file operations and implementation")
Signed-off-by: Raphael Pinsonneault-Thibeault <rpthibeault@...il.com>
---

v1:
https://lore.kernel.org/all/20250925203701.223744-2-rpthibeault@gmail.com/
Differences from v1: 
In the previous version, I thought that mi_read() returning an error was
a genuine reason to stop trying to format the record. That was wrong. If
we find that mi_read() fails and that therefore we cannot reuse an old
record, we should try to make a new one. It then follows that
ntfs_get_bh() should provide an uptodate bh for the new record.

On testing:
I could not get xfstests-bld to work for ntfs3. Config nightmare. So I
first tested with the syzkaller repro, then tested for regressions manually 
with a program that called all the syscall code paths touched by my change: 
mkdir, fallocate (FALLOC_FL_COLLAPSE_RANGE), and 
fallocate (FALLOC_FL_INSERT_RANGE).

 fs/ntfs3/fsntfs.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)

diff --git a/fs/ntfs3/fsntfs.c b/fs/ntfs3/fsntfs.c
index 938d351ebac7..0c07502fdb0f 100644
--- a/fs/ntfs3/fsntfs.c
+++ b/fs/ntfs3/fsntfs.c
@@ -1373,7 +1373,14 @@ int ntfs_get_bh(struct ntfs_sb_info *sbi, const struct runs_tree *run, u64 vbo,
 				}
 				if (buffer_locked(bh))
 					__wait_on_buffer(bh);
-				set_buffer_uptodate(bh);
+
+				lock_buffer(bh);
+				if (!buffer_uptodate(bh))
+				{
+					memset(bh->b_data, 0, blocksize);
+					set_buffer_uptodate(bh);
+				}
+				unlock_buffer(bh);
 			} else {
 				bh = ntfs_bread(sb, block);
 				if (!bh) {
-- 
2.43.0