Re: [PATCH] block: genhd: fix double kfree() in __alloc_disk_node()

[Christoph Hellwig <hch@lst.de>]

On Sun, Sep 19, 2021 at 11:44:29PM +0900, Tetsuo Handa wrote:
> syzbot is reporting use-after-free read at bdev_free_inode() [1], for
> kfree() from __alloc_disk_node() is called before bdev_free_inode()
> (which is called after RCU grace period) reads bdev->bd_disk and calls
> kfree(bdev->bd_disk).
> 
> Fix use-after-free read followed by double kfree() problem
> by explicitly resetting bdev->bd_disk to NULL before calling iput().
> 
> Link: https://syzkaller.appspot.com/bug?extid=8281086e8a6fbfbd952a [1]
> Reported-by: syzbot <syzbot+8281086e8a6fbfbd952a@syzkaller.appspotmail.com>
> Signed-off-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
> ---
> This patch is not tested due to lack of reproducer. Is this fix correct?
> 
>  block/bdev.c  | 1 +
>  block/genhd.c | 1 +
>  2 files changed, 2 insertions(+)
> 
> diff --git a/block/bdev.c b/block/bdev.c
> index cf2780cb44a7..f6b8bac83bd8 100644
> --- a/block/bdev.c
> +++ b/block/bdev.c
> @@ -495,6 +495,7 @@ struct block_device *bdev_alloc(struct gendisk *disk, u8 partno)
>  	bdev->bd_inode = inode;
>  	bdev->bd_stats = alloc_percpu(struct disk_stats);
>  	if (!bdev->bd_stats) {
> +		bdev->bd_disk = NULL;
>  		iput(inode);
>  		return NULL;
>  	}

I was going to suggest to just move the bd_disk initialization after
the bd_stats allocations,  but iseems like we currently don't even
the zero the bdev on allocation.  So I suspect we should do that first
to avoid nasty surprises.