[bpf-next,v1,2/5] bpf: Support getting referenced kptr from struct_ops argument

Commit Message

Amery Hung Feb. 14, 2025, 4:45 p.m. UTC

From: Amery Hung <[e-mail]>

Allows struct_ops programs to acqurie referenced kptrs from arguments
by directly reading the argument.

The verifier will acquire a reference for struct_ops a argument tagged
with "__ref" in the stub function in the beginning of the main program.
The user will be able to access the referenced kptr directly by reading
the context as long as it has not been released by the program.

This new mechanism to acquire referenced kptr (compared to the existing
"kfunc with KF_ACQUIRE") is introduced for ergonomic and semantic reasons.
In the first use case, Qdisc_ops, an skb is passed to .enqueue in the
first argument. This mechanism provides a natural way for users to get a
referenced kptr in the .enqueue struct_ops programs and makes sure that a
qdisc will always enqueue or drop the skb.

Signed-off-by: Amery Hung <[e-mail]>
Acked-by: Eduard Zingerman <[e-mail]>
---
 include/linux/bpf.h         |  3 +++
 kernel/bpf/bpf_struct_ops.c | 26 ++++++++++++++++++++------
 kernel/bpf/btf.c            |  1 +
 kernel/bpf/verifier.c       | 35 ++++++++++++++++++++++++++++++++---
 4 files changed, 56 insertions(+), 9 deletions(-)

Comments

Martin KaFai Lau Feb. 15, 2025, 12:49 a.m. UTC | #1

On 2/14/25 8:45 AM, Amery Hung wrote:
> diff --git a/kernel/bpf/btf.c b/kernel/bpf/btf.c
> index 9de6acddd479..fd3470fbd144 100644
> --- a/kernel/bpf/btf.c
> +++ b/kernel/bpf/btf.c
> @@ -6677,6 +6677,7 @@ bool btf_ctx_access(int off, int size, enum bpf_access_type type,
>   			info->reg_type = ctx_arg_info->reg_type;
>   			info->btf = ctx_arg_info->btf ? : btf_vmlinux;
>   			info->btf_id = ctx_arg_info->btf_id;
> +			info->ref_obj_id = ctx_arg_info->refcounted ? ctx_arg_info->ref_obj_id : 0;

A small nit. No need to check ctx_arg_info->refcounted. If refcounted is false, 
ref_obj_id should have been 0.

			info->ref_obj_id = ctx_arg_info->ref_obj_id;


Acked-by: Martin KaFai Lau <[e-mail]>

Alexei Starovoitov Feb. 15, 2025, 2:48 a.m. UTC | #2

On Fri, Feb 14, 2025 at 8:45 AM Amery Hung <[e-mail]> wrote:
>
> +               if (!is_nullable && !is_refcounted)
>                         continue;
>
> +               if (is_nullable)
> +                       suffix = MAYBE_NULL_SUFFIX;
> +               else if (is_refcounted)
> +                       suffix = REFCOUNTED_SUFFIX;

I would remove the first 'if' and add:
   else
      continue;

>                 /* Should be a pointer to struct */
>                 pointed_type = btf_type_resolve_ptr(btf,
>                                                     args[arg_no].type,
> @@ -236,7 +246,7 @@ static int prepare_arg_info(struct btf *btf,
>                 if (!pointed_type ||
>                     !btf_type_is_struct(pointed_type)) {
>                         pr_warn("stub function %s has %s tagging to an unsupported type\n",
> -                               stub_fname, MAYBE_NULL_SUFFIX);
> +                               stub_fname, suffix);
>                         goto err_out;
>                 }
>
> @@ -254,11 +264,15 @@ static int prepare_arg_info(struct btf *btf,
>                 }
>
>                 /* Fill the information of the new argument */
> -               info->reg_type =
> -                       PTR_TRUSTED | PTR_TO_BTF_ID | PTR_MAYBE_NULL;
>                 info->btf_id = arg_btf_id;
>                 info->btf = btf;
>                 info->offset = offset;
> +               if (is_nullable) {
> +                       info->reg_type = PTR_TRUSTED | PTR_TO_BTF_ID | PTR_MAYBE_NULL;
> +               } else if (is_refcounted) {
> +                       info->reg_type = PTR_TRUSTED | PTR_TO_BTF_ID;
> +                       info->refcounted = true;
> +               }
>
>                 info++;
>                 info_cnt++;
> diff --git a/kernel/bpf/btf.c b/kernel/bpf/btf.c
> index 9de6acddd479..fd3470fbd144 100644
> --- a/kernel/bpf/btf.c
> +++ b/kernel/bpf/btf.c
> @@ -6677,6 +6677,7 @@ bool btf_ctx_access(int off, int size, enum bpf_access_type type,
>                         info->reg_type = ctx_arg_info->reg_type;
>                         info->btf = ctx_arg_info->btf ? : btf_vmlinux;
>                         info->btf_id = ctx_arg_info->btf_id;
> +                       info->ref_obj_id = ctx_arg_info->refcounted ? ctx_arg_info->ref_obj_id : 0;
>                         return true;
>                 }
>         }
> diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
> index a41ba019780f..a0f51903e977 100644
> --- a/kernel/bpf/verifier.c
> +++ b/kernel/bpf/verifier.c
> @@ -1543,6 +1543,17 @@ static void release_reference_state(struct bpf_verifier_state *state, int idx)
>         return;
>  }
>
> +static bool find_reference_state(struct bpf_verifier_state *state, int ptr_id)
> +{
> +       int i;
> +
> +       for (i = 0; i < state->acquired_refs; i++)
> +               if (state->refs[i].id == ptr_id)
> +                       return true;
> +
> +       return false;
> +}
> +
>  static int release_lock_state(struct bpf_verifier_state *state, int type, int id, void *ptr)
>  {
>         int i;
> @@ -5981,7 +5992,8 @@ static int check_packet_access(struct bpf_verifier_env *env, u32 regno, int off,
>  /* check access to 'struct bpf_context' fields.  Supports fixed offsets only */
>  static int check_ctx_access(struct bpf_verifier_env *env, int insn_idx, int off, int size,
>                             enum bpf_access_type t, enum bpf_reg_type *reg_type,
> -                           struct btf **btf, u32 *btf_id, bool *is_retval, bool is_ldsx)
> +                           struct btf **btf, u32 *btf_id, bool *is_retval, bool is_ldsx,
> +                           u32 *ref_obj_id)

It's ok-ish for now, but 11 arguments in a function is pushing it.
We've accumulated enough tech debt here.
Consider a follow up.

Amery Hung Feb. 15, 2025, 4:58 a.m. UTC | #3

On Fri, Feb 14, 2025 at 6:48 PM Alexei Starovoitov
<[e-mail]> wrote:
>
> On Fri, Feb 14, 2025 at 8:45 AM Amery Hung <[e-mail]> wrote:
> >
> > +               if (!is_nullable && !is_refcounted)
> >                         continue;
> >
> > +               if (is_nullable)
> > +                       suffix = MAYBE_NULL_SUFFIX;
> > +               else if (is_refcounted)
> > +                       suffix = REFCOUNTED_SUFFIX;
>
> I would remove the first 'if' and add:
>    else
>       continue;

That looks better than what it is. I will fix it.

>
> >                 /* Should be a pointer to struct */
> >                 pointed_type = btf_type_resolve_ptr(btf,
> >                                                     args[arg_no].type,
> > @@ -236,7 +246,7 @@ static int prepare_arg_info(struct btf *btf,
> >                 if (!pointed_type ||
> >                     !btf_type_is_struct(pointed_type)) {
> >                         pr_warn("stub function %s has %s tagging to an unsupported type\n",
> > -                               stub_fname, MAYBE_NULL_SUFFIX);
> > +                               stub_fname, suffix);
> >                         goto err_out;
> >                 }
> >
> > @@ -254,11 +264,15 @@ static int prepare_arg_info(struct btf *btf,
> >                 }
> >
> >                 /* Fill the information of the new argument */
> > -               info->reg_type =
> > -                       PTR_TRUSTED | PTR_TO_BTF_ID | PTR_MAYBE_NULL;
> >                 info->btf_id = arg_btf_id;
> >                 info->btf = btf;
> >                 info->offset = offset;
> > +               if (is_nullable) {
> > +                       info->reg_type = PTR_TRUSTED | PTR_TO_BTF_ID | PTR_MAYBE_NULL;
> > +               } else if (is_refcounted) {
> > +                       info->reg_type = PTR_TRUSTED | PTR_TO_BTF_ID;
> > +                       info->refcounted = true;
> > +               }
> >
> >                 info++;
> >                 info_cnt++;
> > diff --git a/kernel/bpf/btf.c b/kernel/bpf/btf.c
> > index 9de6acddd479..fd3470fbd144 100644
> > --- a/kernel/bpf/btf.c
> > +++ b/kernel/bpf/btf.c
> > @@ -6677,6 +6677,7 @@ bool btf_ctx_access(int off, int size, enum bpf_access_type type,
> >                         info->reg_type = ctx_arg_info->reg_type;
> >                         info->btf = ctx_arg_info->btf ? : btf_vmlinux;
> >                         info->btf_id = ctx_arg_info->btf_id;
> > +                       info->ref_obj_id = ctx_arg_info->refcounted ? ctx_arg_info->ref_obj_id : 0;
> >                         return true;
> >                 }
> >         }
> > diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
> > index a41ba019780f..a0f51903e977 100644
> > --- a/kernel/bpf/verifier.c
> > +++ b/kernel/bpf/verifier.c
> > @@ -1543,6 +1543,17 @@ static void release_reference_state(struct bpf_verifier_state *state, int idx)
> >         return;
> >  }
> >
> > +static bool find_reference_state(struct bpf_verifier_state *state, int ptr_id)
> > +{
> > +       int i;
> > +
> > +       for (i = 0; i < state->acquired_refs; i++)
> > +               if (state->refs[i].id == ptr_id)
> > +                       return true;
> > +
> > +       return false;
> > +}
> > +
> >  static int release_lock_state(struct bpf_verifier_state *state, int type, int id, void *ptr)
> >  {
> >         int i;
> > @@ -5981,7 +5992,8 @@ static int check_packet_access(struct bpf_verifier_env *env, u32 regno, int off,
> >  /* check access to 'struct bpf_context' fields.  Supports fixed offsets only */
> >  static int check_ctx_access(struct bpf_verifier_env *env, int insn_idx, int off, int size,
> >                             enum bpf_access_type t, enum bpf_reg_type *reg_type,
> > -                           struct btf **btf, u32 *btf_id, bool *is_retval, bool is_ldsx)
> > +                           struct btf **btf, u32 *btf_id, bool *is_retval, bool is_ldsx,
> > +                           u32 *ref_obj_id)
>
> It's ok-ish for now, but 11 arguments in a function is pushing it.
> We've accumulated enough tech debt here.
> Consider a follow up.

Definitely.

A quick look at the check_ctx_access(): Many arguments are used as
input+output at the same time and are part of bpf_insn_access_aux, so
we might as well just let the caller prepare bpf_insn_access_aux. This
should reduce 5 arguments already.

I will send a follow up patch to cleanup this part.

Patch

diff --git a/include/linux/bpf.h b/include/linux/bpf.h
index f4df39e8c735..15164787ce7f 100644
--- a/include/linux/bpf.h
+++ b/include/linux/bpf.h
@@ -968,6 +968,7 @@  struct bpf_insn_access_aux {
 		struct {
 			struct btf *btf;
 			u32 btf_id;
+			u32 ref_obj_id;
 		};
 	};
 	struct bpf_verifier_log *log; /* for verbose logs */
@@ -1481,6 +1482,8 @@  struct bpf_ctx_arg_aux {
 	enum bpf_reg_type reg_type;
 	struct btf *btf;
 	u32 btf_id;
+	u32 ref_obj_id;
+	bool refcounted;
 };
 
 struct btf_mod_pair {
diff --git a/kernel/bpf/bpf_struct_ops.c b/kernel/bpf/bpf_struct_ops.c
index 9b7f3b9c5262..68df8d8b6db3 100644
--- a/kernel/bpf/bpf_struct_ops.c
+++ b/kernel/bpf/bpf_struct_ops.c
@@ -146,6 +146,7 @@  void bpf_struct_ops_image_free(void *image)
 }
 
 #define MAYBE_NULL_SUFFIX "__nullable"
+#define REFCOUNTED_SUFFIX "__ref"
 
 /* Prepare argument info for every nullable argument of a member of a
  * struct_ops type.
@@ -174,11 +175,13 @@  static int prepare_arg_info(struct btf *btf,
 			    struct bpf_struct_ops_arg_info *arg_info)
 {
 	const struct btf_type *stub_func_proto, *pointed_type;
+	bool is_nullable = false, is_refcounted = false;
 	const struct btf_param *stub_args, *args;
 	struct bpf_ctx_arg_aux *info, *info_buf;
 	u32 nargs, arg_no, info_cnt = 0;
 	char ksym[KSYM_SYMBOL_LEN];
 	const char *stub_fname;
+	const char *suffix;
 	s32 stub_func_id;
 	u32 arg_btf_id;
 	int offset;
@@ -223,12 +226,19 @@  static int prepare_arg_info(struct btf *btf,
 	info = info_buf;
 	for (arg_no = 0; arg_no < nargs; arg_no++) {
 		/* Skip arguments that is not suffixed with
-		 * "__nullable".
+		 * "__nullable or __ref".
 		 */
-		if (!btf_param_match_suffix(btf, &stub_args[arg_no],
-					    MAYBE_NULL_SUFFIX))
+		is_nullable = btf_param_match_suffix(btf, &stub_args[arg_no],
+						     MAYBE_NULL_SUFFIX);
+		is_refcounted = btf_param_match_suffix(btf, &stub_args[arg_no],
+						       REFCOUNTED_SUFFIX);
+		if (!is_nullable && !is_refcounted)
 			continue;
 
+		if (is_nullable)
+			suffix = MAYBE_NULL_SUFFIX;
+		else if (is_refcounted)
+			suffix = REFCOUNTED_SUFFIX;
 		/* Should be a pointer to struct */
 		pointed_type = btf_type_resolve_ptr(btf,
 						    args[arg_no].type,
@@ -236,7 +246,7 @@  static int prepare_arg_info(struct btf *btf,
 		if (!pointed_type ||
 		    !btf_type_is_struct(pointed_type)) {
 			pr_warn("stub function %s has %s tagging to an unsupported type\n",
-				stub_fname, MAYBE_NULL_SUFFIX);
+				stub_fname, suffix);
 			goto err_out;
 		}
 
@@ -254,11 +264,15 @@  static int prepare_arg_info(struct btf *btf,
 		}
 
 		/* Fill the information of the new argument */
-		info->reg_type =
-			PTR_TRUSTED | PTR_TO_BTF_ID | PTR_MAYBE_NULL;
 		info->btf_id = arg_btf_id;
 		info->btf = btf;
 		info->offset = offset;
+		if (is_nullable) {
+			info->reg_type = PTR_TRUSTED | PTR_TO_BTF_ID | PTR_MAYBE_NULL;
+		} else if (is_refcounted) {
+			info->reg_type = PTR_TRUSTED | PTR_TO_BTF_ID;
+			info->refcounted = true;
+		}
 
 		info++;
 		info_cnt++;
diff --git a/kernel/bpf/btf.c b/kernel/bpf/btf.c
index 9de6acddd479..fd3470fbd144 100644
--- a/kernel/bpf/btf.c
+++ b/kernel/bpf/btf.c
@@ -6677,6 +6677,7 @@  bool btf_ctx_access(int off, int size, enum bpf_access_type type,
 			info->reg_type = ctx_arg_info->reg_type;
 			info->btf = ctx_arg_info->btf ? : btf_vmlinux;
 			info->btf_id = ctx_arg_info->btf_id;
+			info->ref_obj_id = ctx_arg_info->refcounted ? ctx_arg_info->ref_obj_id : 0;
 			return true;
 		}
 	}
diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
index a41ba019780f..a0f51903e977 100644
--- a/kernel/bpf/verifier.c
+++ b/kernel/bpf/verifier.c
@@ -1543,6 +1543,17 @@  static void release_reference_state(struct bpf_verifier_state *state, int idx)
 	return;
 }
 
+static bool find_reference_state(struct bpf_verifier_state *state, int ptr_id)
+{
+	int i;
+
+	for (i = 0; i < state->acquired_refs; i++)
+		if (state->refs[i].id == ptr_id)
+			return true;
+
+	return false;
+}
+
 static int release_lock_state(struct bpf_verifier_state *state, int type, int id, void *ptr)
 {
 	int i;
@@ -5981,7 +5992,8 @@  static int check_packet_access(struct bpf_verifier_env *env, u32 regno, int off,
 /* check access to 'struct bpf_context' fields.  Supports fixed offsets only */
 static int check_ctx_access(struct bpf_verifier_env *env, int insn_idx, int off, int size,
 			    enum bpf_access_type t, enum bpf_reg_type *reg_type,
-			    struct btf **btf, u32 *btf_id, bool *is_retval, bool is_ldsx)
+			    struct btf **btf, u32 *btf_id, bool *is_retval, bool is_ldsx,
+			    u32 *ref_obj_id)
 {
 	struct bpf_insn_access_aux info = {
 		.reg_type = *reg_type,
@@ -6003,8 +6015,16 @@  static int check_ctx_access(struct bpf_verifier_env *env, int insn_idx, int off,
 		*is_retval = info.is_retval;
 
 		if (base_type(*reg_type) == PTR_TO_BTF_ID) {
+			if (info.ref_obj_id &&
+			    !find_reference_state(env->cur_state, info.ref_obj_id)) {
+				verbose(env, "invalid bpf_context access off=%d. Reference may already be released\n",
+					off);
+				return -EACCES;
+			}
+
 			*btf = info.btf;
 			*btf_id = info.btf_id;
+			*ref_obj_id = info.ref_obj_id;
 		} else {
 			env->insn_aux_data[insn_idx].ctx_field_size = info.ctx_field_size;
 		}
@@ -7367,7 +7387,7 @@  static int check_mem_access(struct bpf_verifier_env *env, int insn_idx, u32 regn
 		struct bpf_retval_range range;
 		enum bpf_reg_type reg_type = SCALAR_VALUE;
 		struct btf *btf = NULL;
-		u32 btf_id = 0;
+		u32 btf_id = 0, ref_obj_id = 0;
 
 		if (t == BPF_WRITE && value_regno >= 0 &&
 		    is_pointer_value(env, value_regno)) {
@@ -7380,7 +7400,7 @@  static int check_mem_access(struct bpf_verifier_env *env, int insn_idx, u32 regn
 			return err;
 
 		err = check_ctx_access(env, insn_idx, off, size, t, &reg_type, &btf,
-				       &btf_id, &is_retval, is_ldsx);
+				       &btf_id, &is_retval, is_ldsx, &ref_obj_id);
 		if (err)
 			verbose_linfo(env, insn_idx, "; ");
 		if (!err && t == BPF_READ && value_regno >= 0) {
@@ -7411,6 +7431,7 @@  static int check_mem_access(struct bpf_verifier_env *env, int insn_idx, u32 regn
 				if (base_type(reg_type) == PTR_TO_BTF_ID) {
 					regs[value_regno].btf = btf;
 					regs[value_regno].btf_id = btf_id;
+					regs[value_regno].ref_obj_id = ref_obj_id;
 				}
 			}
 			regs[value_regno].type = reg_type;
@@ -22148,6 +22169,7 @@  static int do_check_common(struct bpf_verifier_env *env, int subprog)
 {
 	bool pop_log = !(env->log.level & BPF_LOG_LEVEL2);
 	struct bpf_subprog_info *sub = subprog_info(env, subprog);
+	struct bpf_prog_aux *aux = env->prog->aux;
 	struct bpf_verifier_state *state;
 	struct bpf_reg_state *regs;
 	int ret, i;
@@ -22255,6 +22277,13 @@  static int do_check_common(struct bpf_verifier_env *env, int subprog)
 		mark_reg_known_zero(env, regs, BPF_REG_1);
 	}
 
+	/* Acquire references for struct_ops program arguments tagged with "__ref" */
+	if (!subprog && env->prog->type == BPF_PROG_TYPE_STRUCT_OPS) {
+		for (i = 0; i < aux->ctx_arg_info_size; i++)
+			aux->ctx_arg_info[i].ref_obj_id = aux->ctx_arg_info[i].refcounted ?
+							  acquire_reference(env, 0) : 0;
+	}
+
 	ret = do_check(env);
 out:
 	/* check for NULL is necessary, since cur_state can be freed inside