Re: [PATCH bpf-next 2/5] bpf: factor out a helper to prepare trampoline for struct_ops prog

Hi

On 9/30/2021 1:56 AM, Martin KaFai Lau wrote:

On Tue, Sep 28, 2021 at 10:52:25AM +0800, Hou Tao wrote:

quoted

Factor out a helper bpf_prepare_st_ops_prog() to prepare trampoline
for BPF_PROG_TYPE_STRUCT_OPS prog. It will be used by .test_run
callback in following patch.

Thanks for the patches.

Thanks for you review.

This preparation change should be the first patch instead.

Will do.

quoted

diff --git a/include/linux/bpf.h b/include/linux/bpf.h
index 155dfcfb8923..002bbb2c8bc7 100644
--- a/include/linux/bpf.h
+++ b/include/linux/bpf.h

@@ -2224,4 +2224,9 @@ int bpf_bprintf_prepare(char *fmt, u32 fmt_size, const u64 *raw_args,
 			u32 **bin_buf, u32 num_args);
 void bpf_bprintf_cleanup(void);
 
+int bpf_prepare_st_ops_prog(struct bpf_tramp_progs *tprogs,
+			    struct bpf_prog *prog,
+			    const struct btf_func_model *model,
+			    void *image, void *image_end);

Move it under where other bpf_struct_ops_.*() resides in bpf.h.

quoted

+
 #endif /* _LINUX_BPF_H */

diff --git a/kernel/bpf/bpf_struct_ops.c b/kernel/bpf/bpf_struct_ops.c
index 9abcc33f02cf..ec3c25174923 100644
--- a/kernel/bpf/bpf_struct_ops.c
+++ b/kernel/bpf/bpf_struct_ops.c

@@ -312,6 +312,20 @@ static int check_zero_holes(const struct btf_type *t, void *data)
 	return 0;
 }
 
+int bpf_prepare_st_ops_prog(struct bpf_tramp_progs *tprogs,
+			    struct bpf_prog *prog,
+			    const struct btf_func_model *model,
+			    void *image, void *image_end)

The existing struct_ops functions in the kernel now have naming like
bpf_struct_ops_.*().  How about renaming it to
bpf_struct_ops_prepare_trampoline()?

bpf_struct_ops_prepare_trampoline() may be a little long, and it will make
the indentations of its parameters look ugly, so how about
bpf_struct_ops_prep_prog() ?

quoted

+{
+	u32 flags;
+
+	tprogs[BPF_TRAMP_FENTRY].progs[0] = prog;
+	tprogs[BPF_TRAMP_FENTRY].nr_progs = 1;
+	flags = model->ret_size > 0 ? BPF_TRAMP_F_RET_FENTRY_RET : 0;
+	return arch_prepare_bpf_trampoline(NULL, image, image_end,
+					   model, flags, tprogs, NULL);
+}
+
 static int bpf_struct_ops_map_update_elem(struct bpf_map *map, void *key,
 					  void *value, u64 flags)
 {

@@ -368,7 +382,6 @@ static int bpf_struct_ops_map_update_elem(struct bpf_map *map, void *key,
 		const struct btf_type *mtype, *ptype;
 		struct bpf_prog *prog;
 		u32 moff;
-		u32 flags;
 
 		moff = btf_member_bit_offset(t, member) / 8;
 		ptype = btf_type_resolve_ptr(btf_vmlinux, member->type, NULL);

@@ -430,14 +443,9 @@ static int bpf_struct_ops_map_update_elem(struct bpf_map *map, void *key,
 			goto reset_unlock;
 		}
 
-		tprogs[BPF_TRAMP_FENTRY].progs[0] = prog;
-		tprogs[BPF_TRAMP_FENTRY].nr_progs = 1;
-		flags = st_ops->func_models[i].ret_size > 0 ?
-			BPF_TRAMP_F_RET_FENTRY_RET : 0;

This change can't apply to bpf-next now because
commit 356ed64991c6 ("bpf: Handle return value of BPF_PROG_TYPE_STRUCT_OPS prog")
is not pulled into bpf-next yet.  Please mention the dependency
in the cover letter if it is still the case in v2.

Thanks for the reminder. Will do.

quoted

-		err = arch_prepare_bpf_trampoline(NULL, image,
-						  st_map->image + PAGE_SIZE,
-						  &st_ops->func_models[i],
-						  flags, tprogs, NULL);
+		err = bpf_prepare_st_ops_prog(tprogs, prog,
+					      &st_ops->func_models[i],
+					      image, st_map->image + PAGE_SIZE);
 		if (err < 0)
 			goto reset_unlock;
 
-- 
2.29.2

.