Re: [syzbot] [fs?] possible deadlock in quotactl_fd

[Christian Brauner <brauner@kernel.org>]

On Tue, Apr 11, 2023 at 12:55:42PM +0200, Jan Kara wrote:
> On Tue 11-04-23 12:11:52, Christian Brauner wrote:
> > On Mon, Apr 10, 2023 at 11:53:46PM -0700, syzbot wrote:
> > > Hello,
> > > 
> > > syzbot found the following issue on:
> > > 
> > > HEAD commit:    0d3eb744aed4 Merge tag 'urgent-rcu.2023.04.07a' of git://g..
> > > git tree:       upstream
> > > console output: https://syzkaller.appspot.com/x/log.txt?x=11798e4bc80000
> > > kernel config:  https://syzkaller.appspot.com/x/.config?x=c21559e740385326
> > > dashboard link: https://syzkaller.appspot.com/bug?extid=cdcd444e4d3a256ada13
> > > compiler:       gcc (Debian 10.2.1-6) 10.2.1 20210110, GNU ld (GNU Binutils for Debian) 2.35.2
> > > 
> > > Unfortunately, I don't have any reproducer for this issue yet.
> > > 
> > > Downloadable assets:
> > > disk image: https://storage.googleapis.com/syzbot-assets/a02928003efa/disk-0d3eb744.raw.xz
> > > vmlinux: https://storage.googleapis.com/syzbot-assets/7839447005a4/vmlinux-0d3eb744.xz
> > > kernel image: https://storage.googleapis.com/syzbot-assets/d26ab3184148/bzImage-0d3eb744.xz
> > > 
> > > IMPORTANT: if you fix the issue, please add the following tag to the commit:
> > > Reported-by: syzbot+cdcd444e4d3a256ada13@syzkaller.appspotmail.com
> > > 
> > > ======================================================
> > > WARNING: possible circular locking dependency detected
> > > 6.3.0-rc6-syzkaller-00016-g0d3eb744aed4 #0 Not tainted
> > > ------------------------------------------------------
> > > syz-executor.3/11858 is trying to acquire lock:
> > > ffff88802a3bc0e0 (&type->s_umount_key#31){++++}-{3:3}, at: __do_sys_quotactl_fd+0x174/0x3f0 fs/quota/quota.c:997
> > > 
> > > but task is already holding lock:
> > > ffff88802a3bc460 (sb_writers#4){.+.+}-{0:0}, at: __do_sys_quotactl_fd+0xd3/0x3f0 fs/quota/quota.c:990
> > > 
> > > which lock already depends on the new lock.
> > > 
> > > 
> > > the existing dependency chain (in reverse order) is:
> > > 
> > > -> #1 (sb_writers#4){.+.+}-{0:0}:
> > >        percpu_down_read include/linux/percpu-rwsem.h:51 [inline]
> > >        __sb_start_write include/linux/fs.h:1477 [inline]
> > >        sb_start_write include/linux/fs.h:1552 [inline]
> > >        write_mmp_block+0xc4/0x820 fs/ext4/mmp.c:50
> > >        ext4_multi_mount_protect+0x50d/0xac0 fs/ext4/mmp.c:343
> > >        __ext4_remount fs/ext4/super.c:6543 [inline]
> > >        ext4_reconfigure+0x242b/0x2b60 fs/ext4/super.c:6642
> > >        reconfigure_super+0x40c/0xa30 fs/super.c:956
> > >        vfs_fsconfig_locked fs/fsopen.c:254 [inline]
> > >        __do_sys_fsconfig+0xa3a/0xc20 fs/fsopen.c:439
> > >        do_syscall_x64 arch/x86/entry/common.c:50 [inline]
> > >        do_syscall_64+0x39/0xb0 arch/x86/entry/common.c:80
> > >        entry_SYSCALL_64_after_hwframe+0x63/0xcd
> > > 
> > > -> #0 (&type->s_umount_key#31){++++}-{3:3}:
> > >        check_prev_add kernel/locking/lockdep.c:3098 [inline]
> > >        check_prevs_add kernel/locking/lockdep.c:3217 [inline]
> > >        validate_chain kernel/locking/lockdep.c:3832 [inline]
> > >        __lock_acquire+0x2ec7/0x5d40 kernel/locking/lockdep.c:5056
> > >        lock_acquire kernel/locking/lockdep.c:5669 [inline]
> > >        lock_acquire+0x1af/0x520 kernel/locking/lockdep.c:5634
> > >        down_write+0x92/0x200 kernel/locking/rwsem.c:1573
> > >        __do_sys_quotactl_fd+0x174/0x3f0 fs/quota/quota.c:997
> > >        do_syscall_x64 arch/x86/entry/common.c:50 [inline]
> > >        do_syscall_64+0x39/0xb0 arch/x86/entry/common.c:80
> > >        entry_SYSCALL_64_after_hwframe+0x63/0xcd
> > > 
> > > other info that might help us debug this:
> > > 
> > >  Possible unsafe locking scenario:
> > > 
> > >        CPU0                    CPU1
> > >        ----                    ----
> > >   lock(sb_writers#4);
> > >                                lock(&type->s_umount_key#31);
> > >                                lock(sb_writers#4);
> > >   lock(&type->s_umount_key#31);
> > > 
> > >  *** DEADLOCK ***
> > 
> > Hmkay, I understand how this happens, I think:
> > 
> > fsconfig(FSCONFIG_CMD_RECONFIGURE)                      quotactl_fd(Q_QUOTAON/Q_QUOTAOFF/Q_XQUOTAON/Q_XQUOTAOFF)
> > 							-> mnt_want_write(f.file->f_path.mnt);
> > -> down_write(&sb->s_umount);                              -> __sb_start_write(sb, SB_FREEZE_WRITE) 
> > -> reconfigure_super(fc);
> >    -> ext4_multi_mount_protect()
> >       -> __sb_start_write(sb, SB_FREEZE_WRITE)         -> down_write(&sb->s_umount);
> > -> up_write(&sb->s_umount);
> 
> Thanks for having a look!
> 
> > I have to step away from the computer now for a bit but naively it seem
> > that the locking order for quotactl_fd() should be the other way around.
> > 
> > But while I'm here, why does quotactl_fd() take mnt_want_write() but
> > quotactl() doesn't? It seems that if one needs to take it both need to
> > take it.
> 
> Couple of notes here:
> 
> 1) quotactl() handles the filesystem freezing by grabbing the s_umount
> semaphore, checking the superblock freeze state (it cannot change while
> s_umount is held) and proceeding if fs is not frozen. This logic is hidden
> in quotactl_block().
> 
> 2) The proper lock ordering is indeed freeze-protection -> s_umount because
> that is implicitely dictated by how filesystem freezing works. If you grab

Yep.

> s_umount and then try to grab freeze protection, you may effectively wait
> for fs to get unfrozen which cannot happen while s_umount is held as
> thaw_super() needs to grab it.

Yep.

> 
> 3) Hence this could be viewed as ext4 bug that it tries to grab freeze
> protection from remount path. *But* reconfigure_super() actually has:
> 
> 	if (sb->s_writers.frozen != SB_UNFROZEN)
> 		return -EBUSY;

And user_get_super() grabs sb->s_umount which means it's not racy to
check for SB_UNFROZEN. I missed that rushing out the door. :)

> 
> so even ext4 is in fact safe because the filesystem is guaranteed to not be
> frozen during remount. But still we should probably tweak the ext4 code to
> avoid this lockdep warning...

Thanks for that!

Christian