Re: [RFC bpf-next v4 5/6] bpf: Create argument information for nullable arguments.

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

On 2/2/24 2:05 PM, thinker.li@gmail.com wrote:
> From: Kui-Feng Lee <thinker.li@gmail.com>
> 
> Collect argument information from the type information of stub functions to
> mark arguments of BPF struct_ops programs with PTR_MAYBE_NULL if they are
> nullable.  A nullable argument is annotated by suffixing "__nullable" at
> the argument name of stub function.
> 
> For nullable arguments, this patch sets an arg_info to label their reg_type
> with PTR_TO_BTF_ID | PTR_TRUSTED | PTR_MAYBE_NULL. This makes the verifier
> to check programs and ensure that they properly check the pointer. The
> programs should check if the pointer is null before accessing the pointed
> memory.
> 
> The implementer of a struct_ops type should annotate the arguments that can
> be null. The implementer should define a stub function (empty) as a
> placeholder for each defined operator. The name of a stub function should
> be in the pattern "<st_op_type>__<operator name>". For example, for
> test_maybe_null of struct bpf_testmod_ops, it's stub function name should
> be "bpf_testmod_ops__test_maybe_null". You mark an argument nullable by
> suffixing the argument name with "__nullable" at the stub function.
> 
> Since we already has stub functions for kCFI, we just reuse these stub
> functions with the naming convention mentioned earlier. These stub
> functions with the naming convention is only required if there are nullable
> arguments to annotate. For functions having not nullable arguments, stub
> functions are not necessary for the purpose of this patch.
> 
> This patch will prepare a list of struct bpf_ctx_arg_aux, aka arg_info, for
> each member field of a struct_ops type.  "arg_info" will be assigned to
> "prog->aux->ctx_arg_info" of BPF struct_ops programs in
> check_struct_ops_btf_id() so that it can be used by btf_ctx_access() later
> to set reg_type properly for the verifier.

I looked at the high level. Some comments below.

> 
> Signed-off-by: Kui-Feng Lee <thinker.li@gmail.com>
> ---
>   include/linux/bpf.h         |  17 ++++
>   kernel/bpf/bpf_struct_ops.c | 166 ++++++++++++++++++++++++++++++++++--
>   kernel/bpf/btf.c            |  14 +++
>   kernel/bpf/verifier.c       |   6 ++
>   4 files changed, 198 insertions(+), 5 deletions(-)
> 
> diff --git a/include/linux/bpf.h b/include/linux/bpf.h
> index 9a2ee9456989..63ef5cbfd213 100644
> --- a/include/linux/bpf.h
> +++ b/include/linux/bpf.h
> @@ -1709,6 +1709,19 @@ struct bpf_struct_ops {
>   	struct btf_func_model func_models[BPF_STRUCT_OPS_MAX_NR_MEMBERS];
>   };
>   
> +/* Every member of a struct_ops type has an instance even the member is not
> + * an operator (function pointer). The "arg_info" field will be assigned to
> + * prog->aux->arg_info of BPF struct_ops programs to provide the argument
> + * information required by the verifier to verify the program.
> + *
> + * btf_ctx_access() will lookup prog->aux->arg_info to find the
> + * corresponding entry for an given argument.
> + */
> +struct bpf_struct_ops_member_arg_info {
> +	struct bpf_ctx_arg_aux *arg_info;
> +	u32 arg_info_cnt;
> +};
> +
>   struct bpf_struct_ops_desc {
>   	struct bpf_struct_ops *st_ops;
>   
> @@ -1716,6 +1729,10 @@ struct bpf_struct_ops_desc {
>   	const struct btf_type *value_type;
>   	u32 type_id;
>   	u32 value_id;
> +
> +	/* Collection of argument information for each member */
> +	struct bpf_struct_ops_member_arg_info *member_arg_info;
> +	u32 member_arg_info_cnt;
>   };
>   
>   enum bpf_struct_ops_state {
> diff --git a/kernel/bpf/bpf_struct_ops.c b/kernel/bpf/bpf_struct_ops.c
> index f98f580de77a..313f6ceabcf4 100644
> --- a/kernel/bpf/bpf_struct_ops.c
> +++ b/kernel/bpf/bpf_struct_ops.c
> @@ -116,17 +116,148 @@ static bool is_valid_value_type(struct btf *btf, s32 value_id,
>   	return true;
>   }
>   
> +#define MAYBE_NULL_SUFFIX "__nullable"
> +#define MAX_STUB_NAME 128
> +
> +static int match_nullable_suffix(const char *name)
> +{
> +	int suffix_len, len;
> +
> +	if (!name)
> +		return 0;
> +
> +	suffix_len = sizeof(MAYBE_NULL_SUFFIX) - 1;
> +	len = strlen(name);
> +	if (len < suffix_len)
> +		return 0;
> +
> +	return !strcmp(name + len - suffix_len, MAYBE_NULL_SUFFIX);
> +}
> +
> +/* Return the type info of a stub function, if it exists.
> + *
> + * The name of the stub function is made up of the name of the struct_ops
> + * and the name of the function pointer member, separated by "__". For
> + * example, if the struct_ops is named "foo_ops" and the function pointer
> + * member is named "bar", the stub function name would be "foo_ops__bar".
> + */
> +static const  struct btf_type *
> +find_stub_func_proto(struct btf *btf, const char *st_op_name,
> +		     const char *member_name)
> +{
> +	char stub_func_name[MAX_STUB_NAME];
> +	const struct btf_type *t, *func_proto;
> +	s32 btf_id;
> +
> +	snprintf(stub_func_name, MAX_STUB_NAME, "%s__%s",
> +		 st_op_name, member_name);
> +	btf_id = btf_find_by_name_kind(btf, stub_func_name, BTF_KIND_FUNC);
> +	if (btf_id < 0)
> +		return NULL;
> +	t = btf_type_by_id(btf, btf_id);
> +	if (!t)
> +		return NULL;
> +	func_proto = btf_type_by_id(btf, t->type);
> +
> +	return func_proto;
> +}
> +
> +/* Prepare argument info for every nullable argument of a member of a
> + * struct_ops type.
> + *
> + * Create and initialize a list of struct bpf_struct_ops_member_arg_info
> + * according to type info of the arguments of the stub functions. (Check
> + * kCFI for more information about stub functions.)
> + *
> + * Each member in the struct_ops type has a struct
> + * bpf_struct_ops_member_arg_info to provide an array of struct
> + * bpf_ctx_arg_aux, which in turn provides the information that used by the
> + * verifier to check the arguments of the BPF struct_ops program assigned
> + * to the member. Here, we only care about the arguments that are marked as
> + * __nullable.
> + *
> + * The array of struct bpf_ctx_arg_aux is eventually assigned to
> + * prog->aux->ctx_arg_info of BPF struct_ops programs and passed to the
> + * verifier. (See check_struct_ops_btf_id())
> + */
> +static int prepare_arg_info(struct btf *btf,
> +			    const char *st_ops_name,
> +			    const char *member_name,
> +			    struct bpf_struct_ops_member_arg_info *member_arg_info)
> +{
> +	const struct btf_type *stub_func_proto, *ptr_type;
> +	struct bpf_ctx_arg_aux *arg_info, *ai_buf = NULL;
> +	const struct btf_param *args;
> +	u32 nargs, arg_no = 0;
> +	const char *arg_name;
> +	s32 arg_btf_id;
> +
> +	stub_func_proto = find_stub_func_proto(btf, st_ops_name, member_name);
> +	if (!stub_func_proto)
> +		return 0;
> +
> +	nargs = btf_type_vlen(stub_func_proto);
> +	if (nargs > MAX_BPF_FUNC_REG_ARGS) {

Checking MAX_BPF_FUNC_REG_ARGS on the stub_func_proto may not be the right 
check. It should have been done on the origin func_proto (i.e. non-stub) when 
preparing the func_model in btf_distill_func_proto(). Please double check.

If it needs to do sanity check on nargs of stub_func_proto, a better check is to 
ensure the narg of the stub_func_proto is the same as the orig_func_proto 
instead. This discrepancy probably should have been complained by the compiler 
already but does not harm to check (==) here in case the argument type is 
changed and a force cast is used (more below).

> +		pr_warn("Cannot support #%u args in stub func %s_stub_%s\n",
> +			nargs, st_ops_name, member_name);
> +		return -EINVAL;
> +	}
> +
> +	ai_buf = kcalloc(nargs, sizeof(*ai_buf), GFP_KERNEL);
> +	if (!ai_buf)
> +		return -ENOMEM;
> +
> +	args = btf_params(stub_func_proto);
> +	for (arg_no = 0; arg_no < nargs; arg_no++) {
> +		/* Skip arguments that is not suffixed with
> +		 * "__nullable".
> +		 */
> +		arg_name = btf_name_by_offset(btf,
> +					      args[arg_no].name_off);
> +		if (!match_nullable_suffix(arg_name))

I have a question/request.

On top of tagging nullable, can we extend the ctx_arg_info idea here to allow 
changing the pointer type?

In particular, take a stub function in bpf_tcp_ca.c:

static u32 bpf_tcp_ca_ssthresh(struct tcp_sock *tp)
{
         return 0;
}

Instead of the "struct sock *sk" argument as defined in the tcp_congestion_ops, 
the stub function uses "struct tcp_sock *tp'. If we can reuse the ctx_arg_info 
idea here, then it can remove the existing way of changing the pointer type from 
bpf_tcp_ca_is_valid_access.

> +			continue;
> +
> +		/* Should be a pointer to struct, array, scalar, or enum */
> +		ptr_type = btf_type_resolve_ptr(btf, args[arg_no].type,
> +						&arg_btf_id);
> +		if (!ptr_type ||
> +		    (!btf_type_is_struct(ptr_type) &&
> +		     !btf_type_is_array(ptr_type) &&
> +		     !btf_type_is_scalar(ptr_type) &&
> +		     !btf_is_any_enum(ptr_type))) {
> +			kfree(ai_buf);
> +			return -EINVAL;
> +		}
> +
> +		/* Fill the information of the new argument */
> +		arg_info = ai_buf + member_arg_info->arg_info_cnt++;
> +		arg_info->reg_type =
> +			PTR_TRUSTED | PTR_MAYBE_NULL | PTR_TO_BTF_ID;
> +		arg_info->btf_id = arg_btf_id;
> +		arg_info->btf = btf;
> +		arg_info->offset = arg_no * sizeof(u64);

I think for the current struct_ops users should be fine to assume sizeof(u64). 
The current struct_ops users should only have pointer/scalar argument (meaning 
there is no struct passed-by-value argument).

I still think it is better to get it correct for all trampoline supported 
argument here. Take a look at 720e6a435194 ("bpf: Allow struct argument in 
trampoline based programs") and get_ctx_arg_idx(). It may be easier (not sure if 
it is cleaner) to directly store the arg_no into arg_info here but arg_info only 
has offset now. Please think about what could be a cleaner way to do it.

> +	}
> +
> +	if (!member_arg_info->arg_info_cnt)
> +		kfree(ai_buf);
> +	else
> +		member_arg_info->arg_info = ai_buf;
> +
> +	return 0;
> +}
> +
>   int bpf_struct_ops_desc_init(struct bpf_struct_ops_desc *st_ops_desc,
>   			     struct btf *btf,
>   			     struct bpf_verifier_log *log)
>   {
> +	struct bpf_struct_ops_member_arg_info *member_arg_info;
>   	struct bpf_struct_ops *st_ops = st_ops_desc->st_ops;
>   	const struct btf_member *member;
>   	const struct btf_type *t;
>   	s32 type_id, value_id;
>   	char value_name[128];
>   	const char *mname;
> -	int i;
> +	int i, err;
>   
>   	if (strlen(st_ops->name) + VALUE_PREFIX_LEN >=
>   	    sizeof(value_name)) {
> @@ -160,6 +291,11 @@ int bpf_struct_ops_desc_init(struct bpf_struct_ops_desc *st_ops_desc,
>   	if (!is_valid_value_type(btf, value_id, t, value_name))
>   		return -EINVAL;
>   
> +	member_arg_info = kcalloc(btf_type_vlen(t), sizeof(*member_arg_info),
> +				  GFP_KERNEL);
> +	if (!member_arg_info)
> +		return -ENOMEM;
> +
>   	for_each_member(i, t, member) {
>   		const struct btf_type *func_proto;
>   
> @@ -167,13 +303,15 @@ int bpf_struct_ops_desc_init(struct bpf_struct_ops_desc *st_ops_desc,
>   		if (!*mname) {
>   			pr_warn("anon member in struct %s is not supported\n",
>   				st_ops->name);
> -			return -EOPNOTSUPP;
> +			err = -EOPNOTSUPP;
> +			goto errout;
>   		}
>   
>   		if (__btf_member_bitfield_size(t, member)) {
>   			pr_warn("bit field member %s in struct %s is not supported\n",
>   				mname, st_ops->name);
> -			return -EOPNOTSUPP;
> +			err = -EOPNOTSUPP;
> +			goto errout;
>   		}
>   
>   		func_proto = btf_type_resolve_func_ptr(btf,
> @@ -185,14 +323,24 @@ int bpf_struct_ops_desc_init(struct bpf_struct_ops_desc *st_ops_desc,
>   					   &st_ops->func_models[i])) {
>   			pr_warn("Error in parsing func ptr %s in struct %s\n",
>   				mname, st_ops->name);
> -			return -EINVAL;
> +			err = -EINVAL;
> +			goto errout;
>   		}
> +
> +		err = prepare_arg_info(btf, st_ops->name, mname,
> +				       member_arg_info + i);
> +		if (err)
> +			goto errout;
>   	}
>   
> +	st_ops_desc->member_arg_info = member_arg_info;
> +	st_ops_desc->member_arg_info_cnt = btf_type_vlen(t);

It should be the same as btf_type_vlen(st_ops_desc->type). I would avoid this 
duplicated info within the same st_ops_desc.

> +
>   	if (st_ops->init(btf)) {
>   		pr_warn("Error in init bpf_struct_ops %s\n",
>   			st_ops->name);
> -		return -EINVAL;
> +		err = -EINVAL;
> +		goto errout;
>   	}
>   
>   	st_ops_desc->type_id = type_id;
> @@ -201,6 +349,14 @@ int bpf_struct_ops_desc_init(struct bpf_struct_ops_desc *st_ops_desc,
>   	st_ops_desc->value_type = btf_type_by_id(btf, value_id);
>   
>   	return 0;
> +
> +errout:
> +	while (i > 0)
> +		kfree(member_arg_info[--i].arg_info);
> +	kfree(member_arg_info);
> +	st_ops_desc->member_arg_info = NULL;
> +
> +	return err;
>   }
>   
>   static int bpf_struct_ops_map_get_next_key(struct bpf_map *map, void *key,
> diff --git a/kernel/bpf/btf.c b/kernel/bpf/btf.c
> index 20d2160b3db5..fd192f69eb78 100644
> --- a/kernel/bpf/btf.c
> +++ b/kernel/bpf/btf.c
> @@ -1699,6 +1699,20 @@ static void btf_free_struct_meta_tab(struct btf *btf)
>   static void btf_free_struct_ops_tab(struct btf *btf)
>   {
>   	struct btf_struct_ops_tab *tab = btf->struct_ops_tab;
> +	struct bpf_struct_ops_member_arg_info *ma_info;
> +	int i, j;
> +	u32 cnt;
> +
> +	if (tab)
> +		for (i = 0; i < tab->cnt; i++) {
> +			ma_info = tab->ops[i].member_arg_info;
> +			if (ma_info) {
> +				cnt = tab->ops[i].member_arg_info_cnt;
> +				for (j = 0; j < cnt; j++)
> +					kfree(ma_info[j].arg_info);
> +			}
> +			kfree(ma_info);
> +		}
>   
>   	kfree(tab);
>   	btf->struct_ops_tab = NULL;
> diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
> index cd4d780e5400..d1d1c2836bc2 100644
> --- a/kernel/bpf/verifier.c
> +++ b/kernel/bpf/verifier.c
> @@ -20373,6 +20373,12 @@ static int check_struct_ops_btf_id(struct bpf_verifier_env *env)
>   		}
>   	}
>   
> +	/* btf_ctx_access() used this to provide argument type info */
> +	prog->aux->ctx_arg_info =
> +		st_ops_desc->member_arg_info[member_idx].arg_info;
> +	prog->aux->ctx_arg_info_size =
> +		st_ops_desc->member_arg_info[member_idx].arg_info_cnt;
> +
>   	prog->aux->attach_func_proto = func_proto;
>   	prog->aux->attach_func_name = mname;
>   	env->ops = st_ops->verifier_ops;