bpf: Return -ENOTSUPP if callbacks are not allowed in non-JITed programs

[PATCH bpf-next v1] bpf: Return -ENOTSUPP if callbacks are not allowed in non-JITed programs

Posted by Tiezhu Yang 2 years ago

If CONFIG_BPF_JIT_ALWAYS_ON is not set and bpf_jit_enable is 0, there
exist 6 failed tests.

  [root@linux bpf]# echo 0 > /proc/sys/net/core/bpf_jit_enable
  [root@linux bpf]# ./test_verifier | grep FAIL
  #107/p inline simple bpf_loop call FAIL
  #108/p don't inline bpf_loop call, flags non-zero FAIL
  #109/p don't inline bpf_loop call, callback non-constant FAIL
  #110/p bpf_loop_inline and a dead func FAIL
  #111/p bpf_loop_inline stack locations for loop vars FAIL
  #112/p inline bpf_loop call in a big program FAIL
  Summary: 505 PASSED, 266 SKIPPED, 6 FAILED

The test log shows that callbacks are not allowed in non-JITed programs,
interpreter doesn't support them yet, thus these tests should be skipped
if jit is disabled, just return -ENOTSUPP instead of -EINVAL for pseudo
calls in fixup_call_args().

With this patch:

  [root@linux bpf]# echo 0 > /proc/sys/net/core/bpf_jit_enable
  [root@linux bpf]# ./test_verifier | grep FAIL
  Summary: 505 PASSED, 272 SKIPPED, 0 FAILED

Signed-off-by: Tiezhu Yang <yangtiezhu@loongson.cn>
---
 kernel/bpf/verifier.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
index a376eb609c41..1c780a893284 100644
--- a/kernel/bpf/verifier.c
+++ b/kernel/bpf/verifier.c
@@ -19069,7 +19069,7 @@ static int fixup_call_args(struct bpf_verifier_env *env)
 			 * have to be rejected, since interpreter doesn't support them yet.
 			 */
 			verbose(env, "callbacks are not allowed in non-JITed programs\n");
-			return -EINVAL;
+			return -ENOTSUPP;
 		}
 
 		if (!bpf_pseudo_call(insn))
-- 
2.42.0

Re: [PATCH bpf-next v1] bpf: Return -ENOTSUPP if callbacks are not allowed in non-JITed programs

Posted by Eduard Zingerman 1 year, 11 months ago

On Mon, 2023-12-25 at 17:18 +0800, Tiezhu Yang wrote:
> If CONFIG_BPF_JIT_ALWAYS_ON is not set and bpf_jit_enable is 0, there
> exist 6 failed tests.
> 
>   [root@linux bpf]# echo 0 > /proc/sys/net/core/bpf_jit_enable
>   [root@linux bpf]# ./test_verifier | grep FAIL
>   #107/p inline simple bpf_loop call FAIL
>   #108/p don't inline bpf_loop call, flags non-zero FAIL
>   #109/p don't inline bpf_loop call, callback non-constant FAIL
>   #110/p bpf_loop_inline and a dead func FAIL
>   #111/p bpf_loop_inline stack locations for loop vars FAIL
>   #112/p inline bpf_loop call in a big program FAIL
>   Summary: 505 PASSED, 266 SKIPPED, 6 FAILED
> 
> The test log shows that callbacks are not allowed in non-JITed programs,
> interpreter doesn't support them yet, thus these tests should be skipped
> if jit is disabled, just return -ENOTSUPP instead of -EINVAL for pseudo
> calls in fixup_call_args().
> 
> With this patch:
> 
>   [root@linux bpf]# echo 0 > /proc/sys/net/core/bpf_jit_enable
>   [root@linux bpf]# ./test_verifier | grep FAIL
>   Summary: 505 PASSED, 272 SKIPPED, 0 FAILED
> 
> Signed-off-by: Tiezhu Yang <yangtiezhu@loongson.cn>
> ---
>  kernel/bpf/verifier.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
> index a376eb609c41..1c780a893284 100644
> --- a/kernel/bpf/verifier.c
> +++ b/kernel/bpf/verifier.c
> @@ -19069,7 +19069,7 @@ static int fixup_call_args(struct bpf_verifier_env *env)
>  			 * have to be rejected, since interpreter doesn't support them yet.
>  			 */
>  			verbose(env, "callbacks are not allowed in non-JITed programs\n");
> -			return -EINVAL;
> +			return -ENOTSUPP;
>  		}
>  
>  		if (!bpf_pseudo_call(insn))

I agree with this change, however I think that it should be consistent.
Quick and non-exhaustive grepping shows that there are 4 places where
"non-JITed" is used in error messages: in check_map_func_compatibility()
and in fixup_call_args().
All these places currently use -EINVAL and should be updated to -ENOTSUPP,
if this change gets a green light.

If the goal is to merely make test_verifier pass when JIT is disabled
there is a different way:
- test_progs has a global variable 'env.jit_enabled' which is used by
  several tests to decide whether to skip or not;
- loop inlining tests could use similar feature, but unfortunately
  test_verifier does not provide similar functionality;
- test_verifier is sort-of in legacy mode, so instead of porting the
  jit_enabled to test_verifier, I think that loop inlining tests
  should be migrated to test_progs. I can do that some time after [1]
  would be landed.
  
[1] https://lore.kernel.org/bpf/20231223104042.1432300-3-houtao@huaweicloud.com/

Re: [PATCH bpf-next v1] bpf: Return -ENOTSUPP if callbacks are not allowed in non-JITed programs

Posted by John Fastabend 1 year, 11 months ago

Eduard Zingerman wrote:
> On Mon, 2023-12-25 at 17:18 +0800, Tiezhu Yang wrote:
> > If CONFIG_BPF_JIT_ALWAYS_ON is not set and bpf_jit_enable is 0, there
> > exist 6 failed tests.
> > 
> >   [root@linux bpf]# echo 0 > /proc/sys/net/core/bpf_jit_enable
> >   [root@linux bpf]# ./test_verifier | grep FAIL
> >   #107/p inline simple bpf_loop call FAIL
> >   #108/p don't inline bpf_loop call, flags non-zero FAIL
> >   #109/p don't inline bpf_loop call, callback non-constant FAIL
> >   #110/p bpf_loop_inline and a dead func FAIL
> >   #111/p bpf_loop_inline stack locations for loop vars FAIL
> >   #112/p inline bpf_loop call in a big program FAIL
> >   Summary: 505 PASSED, 266 SKIPPED, 6 FAILED
> > 
> > The test log shows that callbacks are not allowed in non-JITed programs,
> > interpreter doesn't support them yet, thus these tests should be skipped
> > if jit is disabled, just return -ENOTSUPP instead of -EINVAL for pseudo
> > calls in fixup_call_args().
> > 
> > With this patch:
> > 
> >   [root@linux bpf]# echo 0 > /proc/sys/net/core/bpf_jit_enable
> >   [root@linux bpf]# ./test_verifier | grep FAIL
> >   Summary: 505 PASSED, 272 SKIPPED, 0 FAILED
> > 
> > Signed-off-by: Tiezhu Yang <yangtiezhu@loongson.cn>
> > ---
> >  kernel/bpf/verifier.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
> > index a376eb609c41..1c780a893284 100644
> > --- a/kernel/bpf/verifier.c
> > +++ b/kernel/bpf/verifier.c
> > @@ -19069,7 +19069,7 @@ static int fixup_call_args(struct bpf_verifier_env *env)
> >  			 * have to be rejected, since interpreter doesn't support them yet.
> >  			 */
> >  			verbose(env, "callbacks are not allowed in non-JITed programs\n");
> > -			return -EINVAL;
> > +			return -ENOTSUPP;
> >  		}
> >  
> >  		if (!bpf_pseudo_call(insn))
> 
> I agree with this change, however I think that it should be consistent.
> Quick and non-exhaustive grepping shows that there are 4 places where
> "non-JITed" is used in error messages: in check_map_func_compatibility()
> and in fixup_call_args().
> All these places currently use -EINVAL and should be updated to -ENOTSUPP,
> if this change gets a green light.

My preference is to just leave them as is unless its a serious
problem. In this case any userspace has likely already figured
out how to handle these errors so I don't think there is much
value in changing things.

> 
> If the goal is to merely make test_verifier pass when JIT is disabled
> there is a different way:
> - test_progs has a global variable 'env.jit_enabled' which is used by
>   several tests to decide whether to skip or not;
> - loop inlining tests could use similar feature, but unfortunately
>   test_verifier does not provide similar functionality;
> - test_verifier is sort-of in legacy mode, so instead of porting the
>   jit_enabled to test_verifier, I think that loop inlining tests
>   should be migrated to test_progs. I can do that some time after [1]
>   would be landed.
>   
> [1] https://lore.kernel.org/bpf/20231223104042.1432300-3-houtao@huaweicloud.com/
>

Re: [PATCH bpf-next v1] bpf: Return -ENOTSUPP if callbacks are not allowed in non-JITed programs

Posted by Tiezhu Yang 1 year, 11 months ago


On 01/03/2024 08:05 AM, John Fastabend wrote:
> Eduard Zingerman wrote:
>> On Mon, 2023-12-25 at 17:18 +0800, Tiezhu Yang wrote:
>>> If CONFIG_BPF_JIT_ALWAYS_ON is not set and bpf_jit_enable is 0, there
>>> exist 6 failed tests.

...

>>> Signed-off-by: Tiezhu Yang <yangtiezhu@loongson.cn>
>>> ---
>>>  kernel/bpf/verifier.c | 2 +-
>>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>>
>>> diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
>>> index a376eb609c41..1c780a893284 100644
>>> --- a/kernel/bpf/verifier.c
>>> +++ b/kernel/bpf/verifier.c
>>> @@ -19069,7 +19069,7 @@ static int fixup_call_args(struct bpf_verifier_env *env)
>>>  			 * have to be rejected, since interpreter doesn't support them yet.
>>>  			 */
>>>  			verbose(env, "callbacks are not allowed in non-JITed programs\n");
>>> -			return -EINVAL;
>>> +			return -ENOTSUPP;
>>>  		}
>>>
>>>  		if (!bpf_pseudo_call(insn))
>>
>> I agree with this change, however I think that it should be consistent.
>> Quick and non-exhaustive grepping shows that there are 4 places where
>> "non-JITed" is used in error messages: in check_map_func_compatibility()
>> and in fixup_call_args().
>> All these places currently use -EINVAL and should be updated to -ENOTSUPP,
>> if this change gets a green light.
>
> My preference is to just leave them as is unless its a serious
> problem. In this case any userspace has likely already figured
> out how to handle these errors so I don't think there is much
> value in changing things.

I am not quite sure whether to ignore this patch, but the state of
this patch is "Changes Requested" [1], I guess I should send v2 as
Eduard suggested.

[1] 
https://patchwork.kernel.org/project/netdevbpf/patch/20231225091830.6094-1-yangtiezhu@loongson.cn/

Thanks,
Tiezhu