On 9/15/23 15:43, Martin KaFai Lau wrote:
On 9/12/23 11:14 PM, thinker.li@xxxxxxxxx wrote:
From: Kui-Feng Lee <thinker.li@xxxxxxxxx>
Move most of code to bpf_struct_ops_init_one() that can be use to
initialize new struct_ops types registered dynamically.
While in RFC, still better to have SOB so that it won't be overlooked in
the future.
---
kernel/bpf/bpf_struct_ops.c | 157 +++++++++++++++++++-----------------
1 file changed, 83 insertions(+), 74 deletions(-)
diff --git a/kernel/bpf/bpf_struct_ops.c b/kernel/bpf/bpf_struct_ops.c
index fdc3e8705a3c..1662875e0ebe 100644
--- a/kernel/bpf/bpf_struct_ops.c
+++ b/kernel/bpf/bpf_struct_ops.c
@@ -110,102 +110,111 @@ const struct bpf_prog_ops
bpf_struct_ops_prog_ops = {
static const struct btf_type *module_type;
-void bpf_struct_ops_init(struct btf *btf, struct bpf_verifier_log *log)
+static void bpf_struct_ops_init_one(struct bpf_struct_ops *st_ops,
+ struct btf *btf,
+ struct bpf_verifier_log *log)
{
- s32 type_id, value_id, module_id;
const struct btf_member *member;
- struct bpf_struct_ops *st_ops;
const struct btf_type *t;
+ s32 type_id, value_id;
char value_name[128];
const char *mname;
- u32 i, j;
+ int i;
- /* Ensure BTF type is emitted for "struct bpf_struct_ops_##_name" */
-#define BPF_STRUCT_OPS_TYPE(_name) BTF_TYPE_EMIT(struct
bpf_struct_ops_##_name);
-#include "bpf_struct_ops_types.h"
-#undef BPF_STRUCT_OPS_TYPE
+ if (strlen(st_ops->name) + VALUE_PREFIX_LEN >=
+ sizeof(value_name)) {
+ pr_warn("struct_ops name %s is too long\n",
+ st_ops->name);
+ return;
+ }
+ sprintf(value_name, "%s%s", VALUE_PREFIX, st_ops->name);
- module_id = btf_find_by_name_kind(btf, "module", BTF_KIND_STRUCT);
- if (module_id < 0) {
- pr_warn("Cannot find struct module in btf_vmlinux\n");
+ value_id = btf_find_by_name_kind(btf, value_name,
+ BTF_KIND_STRUCT);
It needs to do some sanity checks on the value_type since this won't be
statically enforced by bpf_struct_ops.c.
Do you mean to check if a value_type has refcnt, state and data field?
[ ... ]
+void bpf_struct_ops_init(struct btf *btf, struct bpf_verifier_log *log)
+{
+ struct bpf_struct_ops *st_ops;
+ s32 module_id;
+ u32 i;
- if (__btf_member_bitfield_size(t, member)) {
- pr_warn("bit field member %s in struct %s is not
supported\n",
- mname, st_ops->name);
- break;
- }
+ /* Ensure BTF type is emitted for "struct bpf_struct_ops_##_name" */
+#define BPF_STRUCT_OPS_TYPE(_name) BTF_TYPE_EMIT(struct
bpf_struct_ops_##_name);
+#include "bpf_struct_ops_types.h"
+#undef BPF_STRUCT_OPS_TYPE
Can this static way of defining struct_ops be removed? bpf_tcp_ca should
be able to use the register_bpf_struct_ops() introduced in patch 2.
It sounds good for me.
For the future subsystem supporting struct_ops, the subsystem could be
compiled as a kernel module or as a built-in. register_bpf_struct_ops()
should work for both.
