Re: [PATCH] Btrfs: fix divide error upon chunk's stripe_len

[Liu Bo <bo.li.liu@oracle.com>]

On Wed, Apr 27, 2016 at 07:33:18PM +0200, David Sterba wrote:
> On Wed, Apr 27, 2016 at 10:23:35AM -0700, Liu Bo wrote:
> > On Wed, Apr 27, 2016 at 06:39:03PM +0200, David Sterba wrote:
> > > On Tue, Apr 26, 2016 at 05:53:31PM -0700, Liu Bo wrote:
> > > > The struct 'map_lookup' uses type int for @stripe_len, while
> > > > btrfs_chunk_stripe_len() can return a u64 value, and it may end up with
> > > > @stripe_len being undefined value and it can lead to 'divide error' in
> > > >  __btrfs_map_block().
> > > > 
> > > > This changes 'map_lookup' to use type u64 for stripe_len, also right now
> > > > we only use BTRFS_STRIPE_LEN for stripe_len, so this adds a valid checker for
> > > > BTRFS_STRIPE_LEN.
> > > > 
> > > > Reported-by: Vegard Nossum <vegard.nossum@oracle.com>
> > > > Reported-by: Quentin Casasnovas <quentin.casasnovas@oracle.com>
> > > 
> > > I smell some fuzzing :) do you have the image available? I'll add it to
> > > the rest in btrfsprogs.
> > 
> > Sure, it's on the way, I'll send it along with a patch for btrfsck (we
> > have to add the same validation check for superblock and chunk in
> > btrfsck.)
> 
> Great!
> 
> > > > Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
> > > > ---
> > > >  fs/btrfs/volumes.c | 2 +-
> > > >  fs/btrfs/volumes.h | 2 +-
> > > >  2 files changed, 2 insertions(+), 2 deletions(-)
> > > > 
> > > > diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
> > > > index e2b54d5..b5cb859 100644
> > > > --- a/fs/btrfs/volumes.c
> > > > +++ b/fs/btrfs/volumes.c
> > > > @@ -6242,7 +6242,7 @@ static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
> > > >  			"invalid chunk length %llu", length);
> > > >  		return -EIO;
> > > >  	}
> > > > -	if (!is_power_of_2(stripe_len)) {
> > > > +	if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) {

We don't need the first 'is_power_of_2' check.

And I think we may need to have another helper, such as btrfs_check_chunk_valid(),
to cover all these (both current and future) validation checks.  What do you think?

Thanks,

-liubo

> > > 
> > > Unfortunatelly this will break current state, as mkfs does not set the
> > > stripe length to 64k but to 4k. But the value is otherwise ignored in
> > > kernel.
> > 
> > This is chunk's stripe_len, not superblock's stripe_len:
> > 
> > make_btrfs() {
> > 	...
> > 	btrfs_set_super_stripesize(&super, cfg->stripesize); --> 4096
> > 	...
> > 	btrfs_set_chunk_stripe_len(buf, chunk, 64 * 1024);
> > }
> 
> Oh right, and the hardcoded stripe chunk size would need to be fixed in
> a lot more places so it's fine. Consider this
> 
> Reviewed-by: David Sterba <dsterba@suse.com>
> 
> and on the way to for-next.