[RFC PATCH bpf-next] bpf: Check percpu map value size first

[RFC PATCH bpf-next] bpf: Check percpu map value size first

[Tao Chen]

Percpu map is often used, but the map value size limit often ignored,
like issue: https://github.com/iovisor/bcc/issues/2519. Actually,
percpu map value size is bound by PCPU_MIN_UNIT_SZIE, so we
can check the value size whether it exceeds PCPU_MIN_UNIT_SZIE first,
like percpu map of local_storage. Maybe the error message seems clearer
compared with "cannot allocate memory".

the test case we create a percpu map with large value like:
struct test_t {
        int a[100000];
};
struct {
        __uint(type, BPF_MAP_TYPE_PERCPU_ARRAY);
        __uint(max_entries, 1);
        __type(key, u32);
        __type(value, struct test_t);
} start SEC(".maps");

test on ubuntu24.04 vm:
libbpf: Error in bpf_create_map_xattr(start):Argument list too long(-7).
Retrying without BTF.

Signed-off-by: Tao Chen <chen.dylane@gmail.com>
---
 kernel/bpf/arraymap.c | 3 +++
 kernel/bpf/hashtab.c  | 3 +++
 2 files changed, 6 insertions(+)

diff --git a/kernel/bpf/arraymap.c b/kernel/bpf/arraymap.c
index a43e62e2a8bb..7c3c32f156ff 100644
--- a/kernel/bpf/arraymap.c
+++ b/kernel/bpf/arraymap.c
@@ -73,6 +73,9 @@ int array_map_alloc_check(union bpf_attr *attr)
 	/* avoid overflow on round_up(map->value_size) */
 	if (attr->value_size > INT_MAX)
 		return -E2BIG;
+	/* percpu map value size is bound by PCPU_MIN_UNIT_SIZE */
+	if (percpu && attr->value_size > PCPU_MIN_UNIT_SIZE)
+		return -E2BIG;
 
 	return 0;
 }
diff --git a/kernel/bpf/hashtab.c b/kernel/bpf/hashtab.c
index 45c7195b65ba..16d590fe1ffb 100644
--- a/kernel/bpf/hashtab.c
+++ b/kernel/bpf/hashtab.c
@@ -462,6 +462,9 @@ static int htab_map_alloc_check(union bpf_attr *attr)
 		 * kmalloc-able later in htab_map_update_elem()
 		 */
 		return -E2BIG;
+	/* percpu map value size is bound by PCPU_MIN_UNIT_SIZE */
+	if (percpu && attr->value_size > PCPU_MIN_UNIT_SIZE)
+		return -E2BIG;
 
 	return 0;
 }
-- 
2.25.1

Re: [RFC PATCH bpf-next] bpf: Check percpu map value size first

[Hou Tao]

Hi,

On 9/6/2024 1:14 AM, Tao Chen wrote:
> Percpu map is often used, but the map value size limit often ignored,
> like issue: https://github.com/iovisor/bcc/issues/2519. Actually,
> percpu map value size is bound by PCPU_MIN_UNIT_SZIE, so we

s/SZIE/SIZE
> can check the value size whether it exceeds PCPU_MIN_UNIT_SZIE first,

The same typo here.
> like percpu map of local_storage. Maybe the error message seems clearer
> compared with "cannot allocate memory".
>
> the test case we create a percpu map with large value like:
> struct test_t {
>         int a[100000];
> };
> struct {
>         __uint(type, BPF_MAP_TYPE_PERCPU_ARRAY);
>         __uint(max_entries, 1);
>         __type(key, u32);
>         __type(value, struct test_t);
> } start SEC(".maps");
>
> test on ubuntu24.04 vm:
> libbpf: Error in bpf_create_map_xattr(start):Argument list too long(-7).
> Retrying without BTF.

Could you please convert it into a separated bpf selftest patch ?
>
> Signed-off-by: Tao Chen <chen.dylane@gmail.com>
> ---
>  kernel/bpf/arraymap.c | 3 +++
>  kernel/bpf/hashtab.c  | 3 +++
>  2 files changed, 6 insertions(+)
>
> diff --git a/kernel/bpf/arraymap.c b/kernel/bpf/arraymap.c
> index a43e62e2a8bb..7c3c32f156ff 100644
> --- a/kernel/bpf/arraymap.c
> +++ b/kernel/bpf/arraymap.c
> @@ -73,6 +73,9 @@ int array_map_alloc_check(union bpf_attr *attr)
>  	/* avoid overflow on round_up(map->value_size) */
>  	if (attr->value_size > INT_MAX)
>  		return -E2BIG;
> +	/* percpu map value size is bound by PCPU_MIN_UNIT_SIZE */
> +	if (percpu && attr->value_size > PCPU_MIN_UNIT_SIZE)
> +		return -E2BIG;
>  
>  	return 0;
>  }

Make sense. However because the map passes round_up(attr->value_size, 8)
to bpf_map_alloc_percpu(). Is it more reasonable to check
round_up(value_size) > PCPU_MIN_UNIT_SIZE instead ?
> diff --git a/kernel/bpf/hashtab.c b/kernel/bpf/hashtab.c
> index 45c7195b65ba..16d590fe1ffb 100644
> --- a/kernel/bpf/hashtab.c
> +++ b/kernel/bpf/hashtab.c
> @@ -462,6 +462,9 @@ static int htab_map_alloc_check(union bpf_attr *attr)
>  		 * kmalloc-able later in htab_map_update_elem()
>  		 */
>  		return -E2BIG;
> +	/* percpu map value size is bound by PCPU_MIN_UNIT_SIZE */
> +	if (percpu && attr->value_size > PCPU_MIN_UNIT_SIZE)
> +		return -E2BIG;
>  

The percpu allocation logic is the same as the array map:
round_up(value_size, 8) is used.
>  	return 0;
>  }

Re: [RFC PATCH bpf-next] bpf: Check percpu map value size first

[Tao Chen]

在 2024/9/6 11:20, Hou Tao 写道:
> Hi,
> 
> On 9/6/2024 1:14 AM, Tao Chen wrote:
>> Percpu map is often used, but the map value size limit often ignored,
>> like issue: https://github.com/iovisor/bcc/issues/2519. Actually,
>> percpu map value size is bound by PCPU_MIN_UNIT_SZIE, so we
> 
> s/SZIE/SIZE

Hi Hou, thank you for your reply!
My bad, i will fix it in v2.

>> can check the value size whether it exceeds PCPU_MIN_UNIT_SZIE first,
> 
> The same typo here.
>> like percpu map of local_storage. Maybe the error message seems clearer
>> compared with "cannot allocate memory".
>>
>> the test case we create a percpu map with large value like:
>> struct test_t {
>>          int a[100000];
>> };
>> struct {
>>          __uint(type, BPF_MAP_TYPE_PERCPU_ARRAY);
>>          __uint(max_entries, 1);
>>          __type(key, u32);
>>          __type(value, struct test_t);
>> } start SEC(".maps");
>>
>> test on ubuntu24.04 vm:
>> libbpf: Error in bpf_create_map_xattr(start):Argument list too long(-7).
>> Retrying without BTF.
> 
> Could you please convert it into a separated bpf selftest patch ?

No problem, i will add test case in v2.

>>
>> Signed-off-by: Tao Chen <chen.dylane@gmail.com>
>> ---
>>   kernel/bpf/arraymap.c | 3 +++
>>   kernel/bpf/hashtab.c  | 3 +++
>>   2 files changed, 6 insertions(+)
>>
>> diff --git a/kernel/bpf/arraymap.c b/kernel/bpf/arraymap.c
>> index a43e62e2a8bb..7c3c32f156ff 100644
>> --- a/kernel/bpf/arraymap.c
>> +++ b/kernel/bpf/arraymap.c
>> @@ -73,6 +73,9 @@ int array_map_alloc_check(union bpf_attr *attr)
>>   	/* avoid overflow on round_up(map->value_size) */
>>   	if (attr->value_size > INT_MAX)
>>   		return -E2BIG;
>> +	/* percpu map value size is bound by PCPU_MIN_UNIT_SIZE */
>> +	if (percpu && attr->value_size > PCPU_MIN_UNIT_SIZE)
>> +		return -E2BIG;
>>   
>>   	return 0;
>>   }
> 
> Make sense. However because the map passes round_up(attr->value_size, 8)

Yeah, you are right, it seems better, i will add it in v2.

> to bpf_map_alloc_percpu(). Is it more reasonable to check
> round_up(value_size) > PCPU_MIN_UNIT_SIZE instead ?
>> diff --git a/kernel/bpf/hashtab.c b/kernel/bpf/hashtab.c
>> index 45c7195b65ba..16d590fe1ffb 100644
>> --- a/kernel/bpf/hashtab.c
>> +++ b/kernel/bpf/hashtab.c
>> @@ -462,6 +462,9 @@ static int htab_map_alloc_check(union bpf_attr *attr)
>>   		 * kmalloc-able later in htab_map_update_elem()
>>   		 */
>>   		return -E2BIG;
>> +	/* percpu map value size is bound by PCPU_MIN_UNIT_SIZE */
>> +	if (percpu && attr->value_size > PCPU_MIN_UNIT_SIZE)
>> +		return -E2BIG;
>>   
> 
> The percpu allocation logic is the same as the array map:
> round_up(value_size, 8) is used.

ok.

>>   	return 0;
>>   }
> 


-- 
Best Regards
Dylane Chen