On 12/16/23 08:41, Martin KaFai Lau wrote:
On 12/15/23 10:07 PM, Kui-Feng Lee wrote:
On 12/15/23 21:55, Kui-Feng Lee wrote:
On 12/15/23 16:19, Martin KaFai Lau wrote:
On 12/15/23 2:10 PM, Kui-Feng Lee wrote:
On 12/14/23 18:44, Martin KaFai Lau wrote:
On 12/8/23 4:27 PM, thinker.li@xxxxxxxxx wrote:
@@ -681,15 +682,30 @@ static struct bpf_map
*bpf_struct_ops_map_alloc(union bpf_attr *attr)
struct bpf_struct_ops_map *st_map;
const struct btf_type *t, *vt;
struct bpf_map *map;
+ struct btf *btf;
int ret;
- st_ops_desc = bpf_struct_ops_find_value(btf_vmlinux,
attr->btf_vmlinux_value_type_id);
- if (!st_ops_desc)
- return ERR_PTR(-ENOTSUPP);
+ if (attr->value_type_btf_obj_fd) {
+ /* The map holds btf for its whole life time. */
+ btf = btf_get_by_fd(attr->value_type_btf_obj_fd);
+ if (IS_ERR(btf))
+ return ERR_PTR(PTR_ERR(btf));
return ERR_CAST(btf);
It needs to check for btf_is_module:
if (!btf_is_module(btf)) {
btf_put(btf);
return ERR_PTR(-EINVAL);
}
Even btf is btf_vmlinux the kernel's btf, it still works.
btf could be a bpf program's btf. It needs to ensure it is a kernel
module btf here.
Got it!
Isn't btf_is_kernel() better here?
User space may pass a fd to btf_vmlinux.
Limit it to btf_is_module. What is the benefit of supporting btf_vmlinux
as a fd while fd 0 already means btf_vmlinux?
kfunc does not support the btf_vmlinux as fd also, supporting in
struct_ops alone is confusing. I don't think the major user (libbpf)
does this either, so really not much usage other than a confusing API to
have both fd == 0 and a btf_vmlinux's fd to mean btf_vmlinux.
It is fair.
Although libbpf pass 0 as the value of value_type_btf_obj_fd for
btf_vmlinux now, it should be OK for a user space loader to
pass a fd of btf_vmlinux.
WDYT?
