Forwarded: [PATCH] ocfs2: validate i_blkno and BITMAP_FL consistency in ocfs2_validate_inode_block

syzbot posted 1 patch 2 weeks ago
There is a newer version of this series
fs/ocfs2/inode.c | 21 +++++++++++++++++++++
1 file changed, 21 insertions(+)
Forwarded: [PATCH] ocfs2: validate i_blkno and BITMAP_FL consistency in ocfs2_validate_inode_block
Posted by syzbot 2 weeks ago
For archival purposes, forwarding an incoming command email to
linux-kernel@vger.kernel.org, syzkaller-bugs@googlegroups.com.

***

Subject: [PATCH] ocfs2: validate i_blkno and BITMAP_FL consistency in ocfs2_validate_inode_block
Author: kartikey406@gmail.com

#syz test: git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git master

ocfs2_is_cluster_bitmap() checks whether an inode is the global bitmap
by comparing ip_blkno with osb->bitmap_blkno. A corrupted filesystem
can have an inode with i_blkno and OCFS2_BITMAP_FL flag inconsistent:
- i_blkno matches bitmap_blkno but BITMAP_FL is not set, or
- BITMAP_FL is set but i_blkno does not match bitmap_blkno

In either case, ocfs2_is_cluster_bitmap() returns incorrect results,
triggering BUG_ON in ocfs2_block_group_search() or
ocfs2_cluster_group_search() causing a kernel panic.

Call trace:
  ocfs2_block_group_search+0x1c7/0x2c0 fs/ocfs2/suballoc.c:1611
  ocfs2_search_chain+0x38a/0x1010 fs/ocfs2/suballoc.c:1764
  ocfs2_claim_suballoc_bits+0x3a4/0x650 fs/ocfs2/suballoc.c:1978
  ocfs2_claim_new_inode+0x95/0x130 fs/ocfs2/suballoc.c:2137
  ocfs2_mknod_locked+0x129/0x510 fs/ocfs2/namei.c:568
  ocfs2_mknod+0x5c7/0x11d0 fs/ocfs2/namei.c:802
  ocfs2_create+0x136/0x170 fs/ocfs2/namei.c:852

Add validation in ocfs2_validate_inode_block() to check that i_blkno
and OCFS2_BITMAP_FL are consistent:
- If i_blkno matches bitmap_blkno, BITMAP_FL must be set
- If BITMAP_FL is set, i_blkno must match bitmap_blkno

This catches corrupted inodes early when reading from disk.

Reported-by: syzbot+44c564a3cb08605f34a1@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=44c564a3cb08605f34a1
Tested-by: syzbot+44c564a3cb08605f34a1@syzkaller.appspotmail.com
Signed-off-by: Deepanshu Kartikey <kartikey406@gmail.com>
---
v3:
  - Move validation to ocfs2_validate_inode_block() to catch corruption
    early when inode is read from disk (Joseph Qi)
  - Add check for both directions of inconsistency

v2:
  - Fix commit message - ocfs2_is_cluster_bitmap() checks ip_blkno
    against bitmap_blkno, not a flag (Joseph Qi)
---
 fs/ocfs2/inode.c | 21 +++++++++++++++++++++
 1 file changed, 21 insertions(+)

diff --git a/fs/ocfs2/inode.c b/fs/ocfs2/inode.c
index b5fcc2725a29..2d80934d14b4 100644
--- a/fs/ocfs2/inode.c
+++ b/fs/ocfs2/inode.c
@@ -1529,6 +1529,27 @@ int ocfs2_validate_inode_block(struct super_block *sb,
 		}
 	}
 
+	/*
+	 * Validate bitmap flag consistency with bitmap_blkno.
+	 * A corrupted inode with mismatched i_blkno and BITMAP_FL would
+	 * cause ocfs2_is_cluster_bitmap() to return incorrect results,
+	 * triggering BUG_ON in ocfs2_block_group_search() or
+	 * ocfs2_cluster_group_search().
+	 */
+	if (le64_to_cpu(di->i_blkno) == OCFS2_SB(sb)->bitmap_blkno &&
+	    !(di->i_flags & cpu_to_le32(OCFS2_BITMAP_FL))) {
+		rc = ocfs2_error(sb,
+				 "Invalid dinode #%llu: i_blkno matches bitmap but BITMAP flag not set\n",
+				 (unsigned long long)bh->b_blocknr);
+		goto bail;
+	} else if (di->i_flags & cpu_to_le32(OCFS2_BITMAP_FL) &&
+		   le64_to_cpu(di->i_blkno) != OCFS2_SB(sb)->bitmap_blkno) {
+		rc = ocfs2_error(sb,
+				 "Invalid dinode #%llu: BITMAP flag set but i_blkno does not match bitmap\n",
+				 (unsigned long long)bh->b_blocknr);
+		goto bail;
+	}
+
 	rc = 0;
 
 bail:
-- 
2.43.0