On 3.06.20 г. 12:30 ч., Filipe Manana wrote:
> On Wed, Jun 3, 2020 at 8:32 AM Nikolay Borisov <nborisov@xxxxxxxx> wrote:
>>
>>
>>
>> On 1.06.20 г. 21:12 ч., fdmanana@xxxxxxxxxx wrote:
>>> From: Filipe Manana <fdmanana@xxxxxxxx>
>>>
>>> When removing a block group, if we fail to delete the block group's item
>>> from the extent tree, we jump to the 'out' label and end up decrementing
>>> the block group's reference count once only (by 1), resulting in a counter
>>> leak because the block group at that point was already removed from the
>>> block group cache rbtree - so we have to decrement the reference count
>>> twice, once for the rbtree and once for our lookup at the start of the
>>> function.
>>
>> However I'm having hard time reconciling this. The block group is
>> removed from the block_group_cache_tree after we've called
>> btrfs_del_item. So if btrfs_del_item or btrfs_search_slot fail the code
>> jumps at out_put_group and puts the reference acquired at the beginning
>> of the function via btrfs_lookup_block_group.
>>
>> I think what you meant is if we fail to delete the block group's item
>> from the freespace tree, that is if we fail
>> remove_block_group_free_space, then we'd have a ref leak.
>
> What I meant is exactly what I wrote:
> if we fail to delete the block group's item from the extent tree (the
> call to remove_block_group_item()),
> we end up decrementing the reference count only once because we jump
> to the out label - but we
> should have decremented it twice, once for the rbtree removal, which
> happened before, and once for
> the lookup at the start of the function.
Right, and this is case 2 I described in my 2nd email. However my
initial email referred to case 1 from my 2nd email. There are
essentially 2 bugs w.r.t missing a put_block_group: one happens when
remove_block-group_free_space fails and the 2nd one (which you have
described) when remove_block_group_item fails. IMO the change log should
describe the 2 issues.
>
> Thanks.
>
>> With this
>> modification to the changelog:
>>
>> Reviewed-by: Nikolay Borisov <nborisov@xxxxxxxx>
>>
>>>
>>> To make things less error prone, decrement the reference count for the
>>> rbtree immediately after removing the block group from it. This also
>>> eleminates the need for two different exit labels on error, renaming
>>> 'out_put_label' to just 'out' and removing the old 'out'.
>>
>> I agree with this.
>>
>>>
>>> Fixes: f6033c5e333238 ("btrfs: fix block group leak when removing fails")
>>> Signed-off-by: Filipe Manana <fdmanana@xxxxxxxx>
>>
>> <snip>
