For archival purposes, forwarding an incoming command email to
linux-kernel@vger.kernel.org, syzkaller-bugs@googlegroups.com.
***
Subject: [PATCH] nilfs2: fix NULL i_assoc_inode dereference in nilfs_mdt_save_to_shadow_map
Author: kartikey406@gmail.com
#syz test: git://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git master
When nilfs_dat_read() is called a second time during rollback recovery
from a corrupted segment (checksum error), nilfs_iget_locked() returns
a cached DAT inode that is not I_NEW, causing the function to skip
nilfs_mdt_setup_shadow_map(). This leaves i_assoc_inode as NULL in the
DAT inode, which later causes a general protection fault in
nilfs_mdt_save_to_shadow_map() when NILFS_IOCTL_CLEAN_SEGMENTS is
invoked immediately after mount.
Fix this by redirecting the non-I_NEW path to a new reinit_shadow label
that calls nilfs_mdt_setup_shadow_map() if the shadow map has not been
initialized yet, ensuring i_assoc_inode is always valid before the
segment cleaner uses it.
Reported-by: syzbot+4b4093b1f24ad789bf37@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=4b4093b1f24ad789bf37
Signed-off-by: Deepanshu Kartikey <Kartikey406@gmail.com>
---
fs/nilfs2/dat.c | 10 +++++++++-
1 file changed, 9 insertions(+), 1 deletion(-)
diff --git a/fs/nilfs2/dat.c b/fs/nilfs2/dat.c
index 674380837ab9..c0b656e1c4ef 100644
--- a/fs/nilfs2/dat.c
+++ b/fs/nilfs2/dat.c
@@ -507,7 +507,7 @@ int nilfs_dat_read(struct super_block *sb, size_t entry_size,
if (unlikely(!dat))
return -ENOMEM;
if (!(inode_state_read_once(dat) & I_NEW))
- goto out;
+ goto reinit_shadow;
err = nilfs_mdt_init(dat, NILFS_MDT_GFP, sizeof(*di));
if (err)
@@ -529,6 +529,14 @@ int nilfs_dat_read(struct super_block *sb, size_t entry_size,
goto failed;
unlock_new_inode(dat);
+ goto out;
+reinit_shadow:
+ di = NILFS_DAT_I(dat);
+ if (!di->mi.mi_shadow) {
+ err = nilfs_mdt_setup_shadow_map(dat, &di->shadow);
+ if (err)
+ goto failed;
+ }
out:
*inodep = dat;
return 0;
--
2.43.0