Re: [PATCH mm-stable] uprobes: Revert ref_ctr_offset in uprobe_unregister error path

[David Hildenbrand <david@redhat.com>]

On 13.05.25 15:16, David Hildenbrand wrote:
> On 13.05.25 14:21, Jiri Olsa wrote:
>> From: Jiri Olsa <olsajiri@gmail.com>
>>
> 
> Thanks for debugging.
> 
>> There's error path that could lead to inactive uprobe:
>>
>>     1) uprobe_register succeeds - updates instruction to int3 and
>>        changes ref_ctr from 0 to 1
>>     2) uprobe_unregister fails  - int3 stays in place, but ref_ctr
>>        is changed to 0 (it's not restored to 1 in the fail path)
>>        uprobe is leaked
>>     3) another uprobe_register comes and re-uses the leaked uprobe
>>        and succeds - but int3 is already in place, so ref_ctr update
>>        is skipped and it stays 0 - uprobe CAN NOT be triggered now
>>     4) uprobe_unregister fails because ref_ctr value is unexpected
>>
>> Fixing this by reverting the updated ref_ctr value back to 1 in step 2),
>> which is the case when uprobe_unregister fails (int3 stays in place),
>> but we have already updated refctr.
>>
>> The new scenario will go as follows:
>>
>>     1) uprobe_register succeeds - updates instruction to int3 and
>>        changes ref_ctr from 0 to 1
>>     2) uprobe_unregister fails  - int3 stays in place and ref_ctr
>>        is reverted to 1..  uprobe is leaked
>>     3) another uprobe_register comes and re-uses the leaked uprobe
>>        and succeds - but int3 is already in place, so ref_ctr update
>>        is skipped and it stays 1 - uprobe CAN be triggered now
>>     4) uprobe_unregister succeeds
>>
>> Suggested-by: Oleg Nesterov <oleg@redhat.com>
> 
> If it's in mm-stable, we should have
> 
> Fixes: ...
> 
> here
> 
>> Signed-off-by: Jiri Olsa <jolsa@kernel.org>
>> ---
>> Please note it's based on mm-stable branch, because it has the
>> latest uprobe_write_opcode rewrite changes.
>>
>>    kernel/events/uprobes.c | 4 ++--
>>    1 file changed, 2 insertions(+), 2 deletions(-)
>>
>> diff --git a/kernel/events/uprobes.c b/kernel/events/uprobes.c
>> index 4c965ba77f9f..84ee7b590861 100644
>> --- a/kernel/events/uprobes.c
>> +++ b/kernel/events/uprobes.c
>> @@ -581,8 +581,8 @@ int uprobe_write_opcode(struct arch_uprobe *auprobe, struct vm_area_struct *vma,
>>    
>>    out:
>>    	/* Revert back reference counter if instruction update failed. */
>> -	if (ret < 0 && is_register && ref_ctr_updated)
>> -		update_ref_ctr(uprobe, mm, -1);
>> +	if (ret < 0 && ref_ctr_updated)
>> +		update_ref_ctr(uprobe, mm, is_register ? -1 : 1);
> 
> 
> Hm, but my patch essentially did here
> 
>           /* Revert back reference counter if instruction update failed. */
> -       if (ret && is_register && ref_ctr_updated)
> +       if (ret < 0 && is_register && ref_ctr_updated)
>                   update_ref_ctr(uprobe, mm, -1);
> 
> So how come this wasn't a problem before?

Oh, or was this a problem before? Then we should find the corresponding 
commit that needs fixing.

-- 
Cheers,

David / dhildenb