Re: [PATCH bpf-next v3 2/2] bpf: Check return from set_memory_rox()

[Martin KaFai Lau <martin.lau@linux.dev>]

On 3/15/24 2:11 PM, Martin KaFai Lau wrote:
> On 3/15/24 1:55 PM, Martin KaFai Lau wrote:
>> On 3/15/24 10:06 AM, Christophe Leroy wrote:
>>> diff --git a/kernel/bpf/bpf_struct_ops.c b/kernel/bpf/bpf_struct_ops.c
>>> index 43356faaa057..ca1d9b87c475 100644
>>> --- a/kernel/bpf/bpf_struct_ops.c
>>> +++ b/kernel/bpf/bpf_struct_ops.c
>>> @@ -742,8 +742,11 @@ static long bpf_struct_ops_map_update_elem(struct 
>>> bpf_map *map, void *key,
>>>           if (err)
>>>               goto reset_unlock;
>>>       }
>>> -    for (i = 0; i < st_map->image_pages_cnt; i++)
>>> -        arch_protect_bpf_trampoline(st_map->image_pages[i], PAGE_SIZE);
>>> +    for (i = 0; i < st_map->image_pages_cnt && !err; i++)

I was about to apply but I still think checking "&& !err" is not right given how 
"err" is used in the earlier code of this function.

The err may not be 0 in the first iteration of this for loop. Take a look at the 
"if (err > 0)" check in the "for_each_member(i, t, member)" loop above.

>>> +        err = arch_protect_bpf_trampoline(st_map->image_pages[i], PAGE_SIZE);
>>> +
>>> +    if (err)
>>> +        goto reset_unlock;
>>
>> This part does not look right. The "if (err)" check should be inside the for 
>> loop.

Instead of adding an extra "err = 0;" before the for loop. It is better to move 
this "if (err) goto reset_unlock;" into the for loop and remove the "&& !err" 
test above.

> 
> ah. Please ignore. missed the "!err" in the for loop.