ext4: avoid drop reference to iloc.bh twice

ext4: avoid drop reference to iloc.bh twice

[Pan Bian]

The reference to iloc.bh has been dropped in ext4_mark_iloc_dirty.
However, the reference is dropped again if error occurs during
ext4_handle_dirty_metadata, which may result in use-after-free bugs.

Fixes: fb265c9cb49e("ext4: add ext4_sb_bread() to disambiguate ENOMEM
cases")

Signed-off-by: Pan Bian <bianpan2016@163.com>
---
 fs/ext4/resize.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/fs/ext4/resize.c b/fs/ext4/resize.c
index e7ae26e..4d5c0fc 100644
--- a/fs/ext4/resize.c
+++ b/fs/ext4/resize.c
@@ -874,6 +874,7 @@ static int add_new_gdb(handle_t *handle, struct inode *inode,
 	err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
 	if (unlikely(err)) {
 		ext4_std_error(sb, err);
+		iloc.bh = NULL;
 		goto errout;
 	}
 	brelse(dind);
-- 
2.7.4

Re: ext4: avoid drop reference to iloc.bh twice

[Jan Kara]

On Thu 18-04-19 16:31:18, Pan Bian wrote:
> The reference to iloc.bh has been dropped in ext4_mark_iloc_dirty.
> However, the reference is dropped again if error occurs during
> ext4_handle_dirty_metadata, which may result in use-after-free bugs.
> 
> Fixes: fb265c9cb49e("ext4: add ext4_sb_bread() to disambiguate ENOMEM
> cases")
> 
> Signed-off-by: Pan Bian <bianpan2016@163.com>

Thanks for the patch! Good spotting. You can add:

Reviewed-by: Jan Kara <jack@suse.cz>

I'm just wondering: Ted, shouldn't we make ext4_mark_iloc_dirty() clear the
iloc.bh pointer on it's own? I believe this is not the first time we had a
bug like this and it would also catch possible use-after-free issues in
case someone tried to use iloc.bh after the reference has been dropped.

Generally the scheme around ext4_get_inode_loc() and
ext4_mark_iloc_dirty() seems to be somewhat error prone. E.g. a quick audit
shows that there's bh leak in ext4_orphan_del() in one of the error paths.
I'll send patches.

								Honza

> ---
>  fs/ext4/resize.c | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/fs/ext4/resize.c b/fs/ext4/resize.c
> index e7ae26e..4d5c0fc 100644
> --- a/fs/ext4/resize.c
> +++ b/fs/ext4/resize.c
> @@ -874,6 +874,7 @@ static int add_new_gdb(handle_t *handle, struct inode *inode,
>  	err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
>  	if (unlikely(err)) {
>  		ext4_std_error(sb, err);
> +		iloc.bh = NULL;
>  		goto errout;
>  	}
>  	brelse(dind);
> -- 
> 2.7.4
> 
> 
-- 
Jan Kara <jack@suse.com>
SUSE Labs, CR

Re: ext4: avoid drop reference to iloc.bh twice

[Theodore Ts'o]

On Thu, Apr 18, 2019 at 02:35:19PM +0200, Jan Kara wrote:
> On Thu 18-04-19 16:31:18, Pan Bian wrote:
> > The reference to iloc.bh has been dropped in ext4_mark_iloc_dirty.
> > However, the reference is dropped again if error occurs during
> > ext4_handle_dirty_metadata, which may result in use-after-free bugs.
> > 
> > Fixes: fb265c9cb49e("ext4: add ext4_sb_bread() to disambiguate ENOMEM
> > cases")
> > 
> > Signed-off-by: Pan Bian <bianpan2016@163.com>
> 
> Thanks for the patch! Good spotting. You can add:
> 
> Reviewed-by: Jan Kara <jack@suse.cz>

Applied, thanks.

> I'm just wondering: Ted, shouldn't we make ext4_mark_iloc_dirty() clear the
> iloc.bh pointer on it's own? I believe this is not the first time we had a
> bug like this and it would also catch possible use-after-free issues in
> case someone tried to use iloc.bh after the reference has been dropped.
> 
> Generally the scheme around ext4_get_inode_loc() and
> ext4_mark_iloc_dirty() seems to be somewhat error prone. E.g. a quick audit
> shows that there's bh leak in ext4_orphan_del() in one of the error paths.
> I'll send patches.

Good suggestion!

I agree, the interface is error-prone.  Clearing inode.bh afterwards
makes sense.

After we do this, we should also scan the call sites, since there are
some places where we have been calling get_bh(iloc.bh) before-hand, so
that the brelse(iloc.bh) in the cleanup path will work.  Other call
paths add iloc.bh = NULL afterwards so that the brelse() will work
correctly.  So we'll be able to clean up all of this afterwards.

		       	     	     - Ted