diff mbox series

[2/2] block: fix use after free for bd_holder_dir

Message ID 20221103025541.1875809-3-yukuai1@huaweicloud.com
State New
Headers
Series block: bugfix for bd_link_disk_holder() |

Commit Message

Yu Kuai Nov. 3, 2022, 2:55 a.m. UTC 
  From: Yu Kuai <yukuai3@huawei.com>

Currently, the caller of bd_link_disk_holer() get 'bdev' by
blkdev_get_by_dev(), which will look up 'bdev' by inode number 'dev'.
Howerver, it's possible that del_gendisk() can be called currently, and
'bd_holder_dir' can be freed before bd_link_disk_holer() access it, thus
use after free is triggered.

t1:				t2:
bdev = blkdev_get_by_dev
				del_gendisk
				 kobject_put(bd_holder_dir)
				  kobject_free()
bd_link_disk_holder

Fix the problem by checking disk is still live and grabbing a reference
to 'bd_holder_dir' first in bd_link_disk_holder().

Signed-off-by: Yu Kuai <yukuai3@huawei.com>
---
 block/holder.c | 23 ++++++++++++++++-------
 1 file changed, 16 insertions(+), 7 deletions(-)

Comments

Christoph Hellwig Nov. 3, 2022, 8:12 a.m. UTC | #1 
On Thu, Nov 03, 2022 at 10:55:41AM +0800, Yu Kuai wrote:
> From: Yu Kuai <yukuai3@huawei.com>
> 
> Currently, the caller of bd_link_disk_holer() get 'bdev' by
> blkdev_get_by_dev(), which will look up 'bdev' by inode number 'dev'.
> Howerver, it's possible that del_gendisk() can be called currently, and
> 'bd_holder_dir' can be freed before bd_link_disk_holer() access it, thus
> use after free is triggered.
> 
> t1:				t2:
> bdev = blkdev_get_by_dev
> 				del_gendisk
> 				 kobject_put(bd_holder_dir)
> 				  kobject_free()
> bd_link_disk_holder
> 
> Fix the problem by checking disk is still live and grabbing a reference
> to 'bd_holder_dir' first in bd_link_disk_holder().

Looks good with some minor stilistic nipicks:

> +	if (!disk_live(bdev->bd_disk)) {
> +		mutex_unlock(&bdev->bd_disk->open_mutex);
> +		return -ENODEV;
> +	}

This can use a goto out_unlock;

>  		holder->refcnt++;
> +		kobject_put(bdev->bd_holder_dir);
>  		goto out_unlock;
>  	}
>  
>  	holder = kzalloc(sizeof(*holder), GFP_KERNEL);
>  	if (!holder) {
>  		ret = -ENOMEM;
> +		kobject_put(bdev->bd_holder_dir);
>  		goto out_unlock;
>  	}
>  
> @@ -101,16 +114,12 @@ int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
>  		ret = __link_disk_holder(bdev, disk);
>  		if (ret) {
>  			kfree(holder);
> +			kobject_put(bdev->bd_holder_dir);

And I think a goto out_put_holder and out_free_holder would clean this up
nicely.

>  	list_add(&holder->list, &disk->slave_bdevs);
> -	/*
> -	 * del_gendisk drops the initial reference to bd_holder_dir, so we need
> -	 * to keep our own here to allow for cleanup past that point.
> -	 */
> -	kobject_get(bdev->bd_holder_dir);

.. with this then jumping straight to out_unlock.


We should repost a series with my first 7 patches and your two.  I can do
that, but it might take some time as I just got through (minor) knee
surgery and am still at the hospital, so if you have spare cycles feel
free to do it.
Yu Kuai Nov. 3, 2022, 9:45 a.m. UTC | #2 
Hi,

在 2022/11/03 16:12, Christoph Hellwig 写道:
> On Thu, Nov 03, 2022 at 10:55:41AM +0800, Yu Kuai wrote:
>> From: Yu Kuai <yukuai3@huawei.com>
>>
>> Currently, the caller of bd_link_disk_holer() get 'bdev' by
>> blkdev_get_by_dev(), which will look up 'bdev' by inode number 'dev'.
>> Howerver, it's possible that del_gendisk() can be called currently, and
>> 'bd_holder_dir' can be freed before bd_link_disk_holer() access it, thus
>> use after free is triggered.
>>
>> t1:				t2:
>> bdev = blkdev_get_by_dev
>> 				del_gendisk
>> 				 kobject_put(bd_holder_dir)
>> 				  kobject_free()
>> bd_link_disk_holder
>>
>> Fix the problem by checking disk is still live and grabbing a reference
>> to 'bd_holder_dir' first in bd_link_disk_holder().
> 
> Looks good with some minor stilistic nipicks:
> 
>> +	if (!disk_live(bdev->bd_disk)) {
>> +		mutex_unlock(&bdev->bd_disk->open_mutex);
>> +		return -ENODEV;
>> +	}
> 
> This can use a goto out_unlock;

This lock is different from current 'out_unlock', add a new lable will
make the code more complex, I think.
> 
>>   		holder->refcnt++;
>> +		kobject_put(bdev->bd_holder_dir);
>>   		goto out_unlock;
>>   	}
>>   
>>   	holder = kzalloc(sizeof(*holder), GFP_KERNEL);
>>   	if (!holder) {
>>   		ret = -ENOMEM;
>> +		kobject_put(bdev->bd_holder_dir);
>>   		goto out_unlock;
>>   	}
>>   
>> @@ -101,16 +114,12 @@ int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
>>   		ret = __link_disk_holder(bdev, disk);
>>   		if (ret) {
>>   			kfree(holder);
>> +			kobject_put(bdev->bd_holder_dir);
> 
> And I think a goto out_put_holder and out_free_holder would clean this up
> nicely.

Yes, you're right.
> 
>>   	list_add(&holder->list, &disk->slave_bdevs);
>> -	/*
>> -	 * del_gendisk drops the initial reference to bd_holder_dir, so we need
>> -	 * to keep our own here to allow for cleanup past that point.
>> -	 */
>> -	kobject_get(bdev->bd_holder_dir);
> 
> .. with this then jumping straight to out_unlock.

Ok, I'll do that in next version.
> 
> 
> We should repost a series with my first 7 patches and your two.  I can do
> that, but it might take some time as I just got through (minor) knee
> surgery and am still at the hospital, so if you have spare cycles feel
> free to do it.

I'm glad to do that, and have a good rest 😄

Thanks,
Kuai
> 
> .
>
Christoph Hellwig Nov. 3, 2022, 9:57 a.m. UTC | #3 
On Thu, Nov 03, 2022 at 05:45:25PM +0800, Yu Kuai wrote:
> This lock is different from current 'out_unlock', add a new lable will
> make the code more complex, I think.

Of course, same mistake as last time..
diff mbox series

Patch

diff --git a/block/holder.c b/block/holder.c
index 5fc68238ce3a..1c6c5b132a92 100644
--- a/block/holder.c
+++ b/block/holder.c
@@ -78,19 +78,32 @@  int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
 	if (bdev->bd_disk == disk)
 		return -EINVAL;
 
-	mutex_lock(&disk->open_mutex);
+	/*
+	 * del_gendisk drops the initial reference to bd_holder_dir, so we
+	 * need to keep our own here to allow for cleanup past that point.
+	 */
+	mutex_lock(&bdev->bd_disk->open_mutex);
+	if (!disk_live(bdev->bd_disk)) {
+		mutex_unlock(&bdev->bd_disk->open_mutex);
+		return -ENODEV;
+	}
 
-	WARN_ON_ONCE(!bdev->bd_holder);
+	kobject_get(bdev->bd_holder_dir);
+	mutex_unlock(&bdev->bd_disk->open_mutex);
 
+	mutex_lock(&disk->open_mutex);
+	WARN_ON_ONCE(!bdev->bd_holder);
 	holder = bd_find_holder_disk(bdev, disk);
 	if (holder) {
 		holder->refcnt++;
+		kobject_put(bdev->bd_holder_dir);
 		goto out_unlock;
 	}
 
 	holder = kzalloc(sizeof(*holder), GFP_KERNEL);
 	if (!holder) {
 		ret = -ENOMEM;
+		kobject_put(bdev->bd_holder_dir);
 		goto out_unlock;
 	}
 
@@ -101,16 +114,12 @@  int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
 		ret = __link_disk_holder(bdev, disk);
 		if (ret) {
 			kfree(holder);
+			kobject_put(bdev->bd_holder_dir);
 			goto out_unlock;
 		}
 	}
 
 	list_add(&holder->list, &disk->slave_bdevs);
-	/*
-	 * del_gendisk drops the initial reference to bd_holder_dir, so we need
-	 * to keep our own here to allow for cleanup past that point.
-	 */
-	kobject_get(bdev->bd_holder_dir);
 
 out_unlock:
 	mutex_unlock(&disk->open_mutex);