Re: [PATCH] btrfs: don't access possibly stale fs_info data in device_list_add

[David Sterba]

On Fri, Mar 04, 2022 at 07:53:27AM +0800, Anand Jain wrote:
> On 04/03/2022 02:24, David Sterba wrote:
> > On Thu, Mar 03, 2022 at 10:40:27PM +0800, Dongliang Mu wrote:
> >>
> >> Fix this by modifying device->fs_info to NULL too.
> >>
> >> Reported-and-tested-by: syzbot+82650a4e0ed38f218363@xxxxxxxxxxxxxxxxxxxxxxxxx
> >> Signed-off-by: Dongliang Mu <mudongliangabcd@xxxxxxxxx>
> >> ---
> >>   fs/btrfs/volumes.c | 2 +-
> >>   1 file changed, 1 insertion(+), 1 deletion(-)
> >>
> >> diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
> >> index b07d382d53a8..c1325bdae9a1 100644
> >> --- a/fs/btrfs/volumes.c
> >> +++ b/fs/btrfs/volumes.c
> >> @@ -954,7 +954,7 @@ static noinline struct btrfs_device *device_list_add(const char *path,
> >>   						  task_pid_nr(current));
> >>   				return ERR_PTR(-EEXIST);
> >>   			}
> >> -			btrfs_info_in_rcu(device->fs_info,
> >> +			btrfs_info_in_rcu(NULL,
> > 
> > A few lines above this is also NULL and was fixed by 0697d9a61099
> > ("btrfs: don't access possibly stale fs_info data for printing duplicate
> > device"), so yeah we probably need the same here.
> 
> So it appears that device->fs_info was garbage instead of NULL OR
> fs_info->sb was NULL?

I think it's a warning that something could happen, in this case
potential garbage value of fs_info.

> Because we always had a check if fs_info is null in btrfs_printk()
> further the commit a0f6d924cada ("btrfs: remove stub device info from
> messages when we have no fs_info") made it better.

Yeah, that's removing a potential crash but still the NULL value could
come from a freed memory. Seems taht we can't rely on fs_info in
device_list_add at all and passing NULL is the only safe way.