[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <20230629042839.GK8954@mit.edu>
Date: Thu, 29 Jun 2023 00:28:39 -0400
From: "Theodore Ts'o" <tytso@....edu>
To: syzbot <syzbot+5407ecf3112f882d2ef3@...kaller.appspotmail.com>
Cc: adilger.kernel@...ger.ca, linux-ext4@...r.kernel.org,
linux-fsdevel@...r.kernel.org, linux-kernel@...r.kernel.org,
reiserfs-devel@...r.kernel.org, syzkaller-bugs@...glegroups.com
Subject: Re: [syzbot] [ext4?] KASAN: slab-use-after-free Read in __ext4_iget
#syz set subsystems: fs, reiserfs
On Tue, Jun 06, 2023 at 05:11:06PM -0700, syzbot wrote:
> Hello,
>
> syzbot found the following issue on:
>
> HEAD commit: a4d7d7011219 Merge tag 'spi-fix-v6.4-rc5' of git://git.ker..
> git tree: upstream
> console output: https://syzkaller.appspot.com/x/log.txt?x=1455f745280000
> kernel config: https://syzkaller.appspot.com/x/.config?x=7474de833c217bf4
> dashboard link: https://syzkaller.appspot.com/bug?extid=5407ecf3112f882d2ef3
> compiler: Debian clang version 15.0.7, GNU ld (GNU Binutils for Debian) 2.35.2
The stack traces on this are... intersting. The use-after free is
coming when ext4_fill_super() tries to get the root inode, via
iget_locked(sb, EXT2_ROOT_INO)
> BUG: KASAN: slab-use-after-free in __ext4_iget+0x2f2/0x3f30 fs/ext4/inode.c:4700
> Read of size 8 at addr ffff888078ca5550 by task syz-executor.5/26112
...
> __ext4_iget+0x2f2/0x3f30 fs/ext4/inode.c:4700
> __ext4_fill_super fs/ext4/super.c:5446 [inline]
> ext4_fill_super+0x545b/0x6c60 fs/ext4/super.c:5672
However, we are getting back an object which is freed, and which was
originally allocated by reiserfs(!):
> Allocated by task 20729:
> kasan_save_stack mm/kasan/common.c:45 [inline]
> kasan_set_track+0x4f/0x70 mm/kasan/common.c:52
> __kasan_slab_alloc+0x66/0x70 mm/kasan/common.c:328
> kasan_slab_alloc include/linux/kasan.h:186 [inline]
> slab_post_alloc_hook+0x68/0x3a0 mm/slab.h:711
> slab_alloc_node mm/slub.c:3451 [inline]
> slab_alloc mm/slub.c:3459 [inline]
> __kmem_cache_alloc_lru mm/slub.c:3466 [inline]
> kmem_cache_alloc_lru+0x11f/0x2e0 mm/slub.c:3482
> alloc_inode_sb include/linux/fs.h:2705 [inline]
> reiserfs_alloc_inode+0x2a/0xc0 fs/reiserfs/super.c:642
> alloc_inode fs/inode.c:260 [inline]
> iget5_locked+0xa0/0x270 fs/inode.c:1241
> reiserfs_fill_super+0x12e4/0x2620 fs/reiserfs/super.c:2053
> mount_bdev+0x2d0/0x3f0 fs/super.c:1380
There is no reproducer, but it seems to be triggering quite frequently
(over once day --- 20 times since June 16, 2023 as of this writing).
I've checked a number of the reports in the Syzkaller dashboard, and
they are all quite similar; somehow ext4 is getting an inode which is
freed, and whose memory was originally allocated by reiserfs.
I'm not sure if this is a reiserfs bug or a core VFS bug, since this
seems to imply that an an old reiserfs inode was left on the
inode_hashtable when a reiserfs file system was unmounted, and then
the struct super was reused and returned for a fresh ext4 mount, and
then when ext4 tried do an iget_locked(), it got the reserifs inode.
That reiserfs inode was either freed and left on the inode_hashtable,
or lifetime of the reiserfs root inode was allowed to last longer than
the reiserfs superblock (maybe someone is playing RCU games?) and so
since it was left on the inode_hashtable, the attempt to free reiserfs
root inode raced with ext4's attempt to fetch the ext4 inode via
iget_locked().
Perhaps one of the VFS or reiserfs maintainers could take a look?
- Ted
Powered by blists - more mailing lists