bpf: Restrict bpf_sys_bpf to CAP_PERFMON

The verifier cannot perform sufficient validation of any pointers passed
into bpf_attr and treats them as integers rather than pointers. The helper
will then read from arbitrary pointers passed into it. Restrict the helper
to CAP_PERFMON since the security model in BPF of arbitrary kernel read is
CAP_BPF + CAP_PERFMON.

Fixes: af2ac3e ("bpf: Prepare bpf syscall to be used from kernel and user space.")
Signed-off-by: YiFei Zhu <[email protected]>
Signed-off-by: Daniel Borkmann <[email protected]>
Acked-by: Alexei Starovoitov <[email protected]>
Link: https://lore.kernel.org/bpf/[email protected]

Author: zhuyifei1999

kernel/bpf/syscall.c

@@ -5197,7 +5197,7 @@ syscall_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
 {
 	switch (func_id) {
 	case BPF_FUNC_sys_bpf:
-		return &bpf_sys_bpf_proto;
+		return !perfmon_capable() ? NULL : &bpf_sys_bpf_proto;
 	case BPF_FUNC_btf_find_by_name_kind:
 		return &bpf_btf_find_by_name_kind_proto;
 	case BPF_FUNC_sys_close: