Re: [PATCH v2 7/7] btrfs: do not take the device_list_mutex in clone_fs_devices

[Anand Jain <anand.jain@oracle.com>]

On 28/07/2021 05:01, Josef Bacik wrote:
> I got the following lockdep splat while testing seed devices
> 
> ======================================================
> WARNING: possible circular locking dependency detected
> 5.14.0-rc2+ #409 Not tainted
> ------------------------------------------------------
> mount/34004 is trying to acquire lock:
> ffff9eaac48188e0 (&fs_devs->device_list_mutex){+.+.}-{3:3}, at: clone_fs_devices+0x4d/0x170
> 
> but task is already holding lock:
> ffff9eaac766d438 (btrfs-chunk-00){++++}-{3:3}, at: __btrfs_tree_read_lock+0x24/0x100
> 
> which lock already depends on the new lock.
> 
> the existing dependency chain (in reverse order) is:
> 
> -> #2 (btrfs-chunk-00){++++}-{3:3}:
>         down_read_nested+0x46/0x60
>         __btrfs_tree_read_lock+0x24/0x100
>         btrfs_read_lock_root_node+0x31/0x40
>         btrfs_search_slot+0x480/0x930
>         btrfs_update_device+0x63/0x180
>         btrfs_chunk_alloc_add_chunk_item+0xdc/0x3a0
>         btrfs_chunk_alloc+0x281/0x540
>         find_free_extent+0x10ca/0x1790
>         btrfs_reserve_extent+0xbf/0x1d0
>         btrfs_alloc_tree_block+0xb1/0x320
>         __btrfs_cow_block+0x136/0x5f0
>         btrfs_cow_block+0x107/0x210
>         btrfs_search_slot+0x56a/0x930
>         btrfs_truncate_inode_items+0x187/0xef0
>         btrfs_truncate_free_space_cache+0x11c/0x210
>         delete_block_group_cache+0x6f/0xb0
>         btrfs_relocate_block_group+0xf8/0x350
>         btrfs_relocate_chunk+0x38/0x120
>         btrfs_balance+0x79b/0xf00
>         btrfs_ioctl_balance+0x327/0x400
>         __x64_sys_ioctl+0x80/0xb0
>         do_syscall_64+0x38/0x90
>         entry_SYSCALL_64_after_hwframe+0x44/0xae
> 
> -> #1 (&fs_info->chunk_mutex){+.+.}-{3:3}:
>         __mutex_lock+0x7d/0x750
>         btrfs_init_new_device+0x6d6/0x1540
>         btrfs_ioctl+0x1b12/0x2d30
>         __x64_sys_ioctl+0x80/0xb0
>         do_syscall_64+0x38/0x90
>         entry_SYSCALL_64_after_hwframe+0x44/0xae
> 
> -> #0 (&fs_devs->device_list_mutex){+.+.}-{3:3}:
>         __lock_acquire+0x10ea/0x1d90
>         lock_acquire+0xb5/0x2b0
>         __mutex_lock+0x7d/0x750
>         clone_fs_devices+0x4d/0x170
>         btrfs_read_chunk_tree+0x32f/0x800
>         open_ctree+0xae3/0x16f0
>         btrfs_mount_root.cold+0x12/0xea
>         legacy_get_tree+0x2d/0x50
>         vfs_get_tree+0x25/0xc0
>         vfs_kern_mount.part.0+0x71/0xb0
>         btrfs_mount+0x10d/0x380
>         legacy_get_tree+0x2d/0x50
>         vfs_get_tree+0x25/0xc0
>         path_mount+0x433/0xb60
>         __x64_sys_mount+0xe3/0x120
>         do_syscall_64+0x38/0x90
>         entry_SYSCALL_64_after_hwframe+0x44/0xae
> 
> other info that might help us debug this:
> 
> Chain exists of:
>    &fs_devs->device_list_mutex --> &fs_info->chunk_mutex --> btrfs-chunk-00
> 
>   Possible unsafe locking scenario:
> 
>         CPU0                    CPU1
>         ----                    ----
>    lock(btrfs-chunk-00);
>                                 lock(&fs_info->chunk_mutex);
>                                 lock(btrfs-chunk-00);
>    lock(&fs_devs->device_list_mutex);
> 
>   *** DEADLOCK ***
> 
> 3 locks held by mount/34004:
>   #0: ffff9eaad75c00e0 (&type->s_umount_key#47/1){+.+.}-{3:3}, at: alloc_super+0xd5/0x3b0
>   #1: ffffffffbd2dcf08 (uuid_mutex){+.+.}-{3:3}, at: btrfs_read_chunk_tree+0x59/0x800
>   #2: ffff9eaac766d438 (btrfs-chunk-00){++++}-{3:3}, at: __btrfs_tree_read_lock+0x24/0x100
> 
> stack backtrace:
> CPU: 0 PID: 34004 Comm: mount Not tainted 5.14.0-rc2+ #409
> Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.13.0-2.fc32 04/01/2014
> Call Trace:
>   dump_stack_lvl+0x57/0x72
>   check_noncircular+0xcf/0xf0
>   __lock_acquire+0x10ea/0x1d90
>   lock_acquire+0xb5/0x2b0
>   ? clone_fs_devices+0x4d/0x170
>   ? lock_is_held_type+0xa5/0x120
>   __mutex_lock+0x7d/0x750
>   ? clone_fs_devices+0x4d/0x170
>   ? clone_fs_devices+0x4d/0x170
>   ? lockdep_init_map_type+0x47/0x220
>   ? debug_mutex_init+0x33/0x40
>   clone_fs_devices+0x4d/0x170
>   ? lock_is_held_type+0xa5/0x120
>   btrfs_read_chunk_tree+0x32f/0x800
>   ? find_held_lock+0x2b/0x80
>   open_ctree+0xae3/0x16f0
>   btrfs_mount_root.cold+0x12/0xea
>   ? rcu_read_lock_sched_held+0x3f/0x80
>   ? kfree+0x1f6/0x410
>   legacy_get_tree+0x2d/0x50
>   vfs_get_tree+0x25/0xc0
>   vfs_kern_mount.part.0+0x71/0xb0
>   btrfs_mount+0x10d/0x380
>   ? kfree+0x1f6/0x410
>   legacy_get_tree+0x2d/0x50
>   vfs_get_tree+0x25/0xc0
>   path_mount+0x433/0xb60
>   __x64_sys_mount+0xe3/0x120
>   do_syscall_64+0x38/0x90
>   entry_SYSCALL_64_after_hwframe+0x44/0xae
> RIP: 0033:0x7f6cbcd9788e
> 
> It is because we take the ->device_list_mutex in this path while holding
> onto the tree locks in the chunk root.  However we do not need the lock
> here, because we're already holding onto the uuid_mutex, and in fact
> have removed all other uses of the ->device_list_mutex in this path
> because of this.  Remove the ->device_list_mutex locking here, add an
> assert for the uuid_mutex and the problem is fixed.
> 
> Signed-off-by: Josef Bacik <josef@toxicpanda.com>
> ---
>   fs/btrfs/volumes.c | 5 ++---
>   1 file changed, 2 insertions(+), 3 deletions(-)
> 
> diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
> index f622e93a6ff1..bdfcc35335c3 100644
> --- a/fs/btrfs/volumes.c
> +++ b/fs/btrfs/volumes.c
> @@ -1000,11 +1000,12 @@ static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
>   	struct btrfs_device *orig_dev;
>   	int ret = 0;
>   
> +	lockdep_assert_held(&uuid_mutex);
> +
>   	fs_devices = alloc_fs_devices(orig->fsid, NULL);
>   	if (IS_ERR(fs_devices))
>   		return fs_devices;
>   
> -	mutex_lock(&orig->device_list_mutex);
>   	fs_devices->total_devices = orig->total_devices;
>   
>   	list_for_each_entry(orig_dev, &orig->devices, dev_list) {
> @@ -1036,10 +1037,8 @@ static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
>   		device->fs_devices = fs_devices;
>   		fs_devices->num_devices++;
>   	}
> -	mutex_unlock(&orig->device_list_mutex);
>   	return fs_devices;
>   error:
> -	mutex_unlock(&orig->device_list_mutex);
>   	free_fs_devices(fs_devices);
>   	return ERR_PTR(ret);
>   }
> 


  This fix is same as in [1]

  [1]
 
https://patchwork.kernel.org/project/linux-btrfs/patch/23a8830f3be500995e74b45f18862e67c0634c3d.1614793362.git.anand.jain@oracle.com/