Re: [PATCH bpf-next] bpf: add a new bpf argument type ARG_CONST_MAP_PTR_OR_NULL

[Hangbin Liu <liuhangbin@gmail.com>]

On Thu, Jul 16, 2020 at 12:28:16AM +0200, Daniel Borkmann wrote:
> On 7/15/20 9:00 AM, Hangbin Liu wrote:
> > Add a new bpf argument type ARG_CONST_MAP_PTR_OR_NULL which could be
> > used when we want to allow NULL pointer for map parameter. The bpf helper
> > need to take care and check if the map is NULL when use this type.
> > 
> > Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
> 
> Is this patch to be merged into the set in [0] for passing NULL ex_map as discussed?
> Seems you sent out two incomplete sets?
> 
>   [0] https://lore.kernel.org/bpf/20200709013008.3900892-1-liuhangbin@gmail.com/T/#m99a8fa8ffe79d5f00d305c0800ad3abe619294f2

Yes, I did it by intend. I thought these two should be consider as
different feature. So I'd prefer post them separately. Once both
patches are merged, I will post a followup patch to add the NULL pointer
support to xdp multicast helper.

> 
> > ---
> >   include/linux/bpf.h   |  1 +
> >   kernel/bpf/verifier.c | 11 ++++++++---
> >   2 files changed, 9 insertions(+), 3 deletions(-)
> > 
> > diff --git a/include/linux/bpf.h b/include/linux/bpf.h
> > index c67c88ad35f8..9d4dbef3c943 100644
> > --- a/include/linux/bpf.h
> > +++ b/include/linux/bpf.h
> > @@ -253,6 +253,7 @@ enum bpf_arg_type {
> >   	ARG_PTR_TO_ALLOC_MEM,	/* pointer to dynamically allocated memory */
> >   	ARG_PTR_TO_ALLOC_MEM_OR_NULL,	/* pointer to dynamically allocated memory or NULL */
> >   	ARG_CONST_ALLOC_SIZE_OR_ZERO,	/* number of allocated bytes requested */
> > +	ARG_CONST_MAP_PTR_OR_NULL,	/* const argument used as pointer to bpf_map or NULL */
> >   };
> >   /* type of values returned from helper functions */
> > diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
> > index 3c1efc9d08fd..d3551a19853a 100644
> > --- a/kernel/bpf/verifier.c
> > +++ b/kernel/bpf/verifier.c
> > @@ -3849,9 +3849,13 @@ static int check_func_arg(struct bpf_verifier_env *env, u32 arg,
> >   		expected_type = SCALAR_VALUE;
> >   		if (type != expected_type)
> >   			goto err_type;
> > -	} else if (arg_type == ARG_CONST_MAP_PTR) {
> > +	} else if (arg_type == ARG_CONST_MAP_PTR ||
> > +		   arg_type == ARG_CONST_MAP_PTR_OR_NULL) {
> >   		expected_type = CONST_PTR_TO_MAP;
> > -		if (type != expected_type)
> > +		if (register_is_null(reg) &&
> > +		    arg_type == ARG_CONST_MAP_PTR_OR_NULL)
> > +			/* final test in check_stack_boundary() */;
> > +		else if (type != expected_type)
> >   			goto err_type;
> >   	} else if (arg_type == ARG_PTR_TO_CTX ||
> >   		   arg_type == ARG_PTR_TO_CTX_OR_NULL) {
> > @@ -3957,7 +3961,8 @@ static int check_func_arg(struct bpf_verifier_env *env, u32 arg,
> >   		return -EFAULT;
> >   	}
> > -	if (arg_type == ARG_CONST_MAP_PTR) {
> > +	if (arg_type == ARG_CONST_MAP_PTR ||
> > +	    (arg_type == ARG_CONST_MAP_PTR_OR_NULL && !register_is_null(reg))) {
> >   		/* bpf_map_xxx(map_ptr) call: remember that map_ptr */
> >   		meta->map_ptr = reg->map_ptr;
> 
> I would probably have the semantics a bit different in the sense that I would
> update meta->map_ptr to the last ARG_CONST_MAP_PTR, meaning:
> 
>     meta->map_ptr = register_is_null(reg) ? NULL : reg->map_ptr;

Thanks for the suggestion. I will update it.

> 
> >   	} else if (arg_type == ARG_PTR_TO_MAP_KEY) {
> > 
> 
> In combination with the set, this also needs test_verifier selftests in order to
> exercise BPF insn snippets for the good & [expected] bad case.

OK, I will add a test for this.

Thanks
Hangbin