On Fri, Jan 16, 2026 at 5:58 AM Leon Hwang [off-list ref] wrote:

On 2026/1/16 08:42, Andrii Nakryiko wrote:

quoted

On Mon, Jan 12, 2026 at 6:58 AM Leon Hwang [off-list ref] wrote:

quoted

To support the extended BPF syscall introduced in the previous commit,
introduce the following internal APIs:

* 'sys_bpf_ext()'
* 'sys_bpf_ext_fd()'
  They wrap the raw 'syscall()' interface to support passing extended
  attributes.
* 'probe_sys_bpf_ext()'
  Check whether current kernel supports the extended attributes.

Signed-off-by: Leon Hwang <redacted>
---
 tools/lib/bpf/bpf.c             | 34 +++++++++++++++++++++++++++++++++
 tools/lib/bpf/features.c        |  8 ++++++++
 tools/lib/bpf/libbpf_internal.h |  3 +++
 3 files changed, 45 insertions(+)

diff --git a/tools/lib/bpf/bpf.c b/tools/lib/bpf/bpf.c
index 21b57a629916..d44e667aaf02 100644
--- a/tools/lib/bpf/bpf.c
+++ b/tools/lib/bpf/bpf.c

@@ -69,6 +69,40 @@ static inline __u64 ptr_to_u64(const void *ptr)
        return (__u64) (unsigned long) ptr;
 }

+static inline int sys_bpf_ext(enum bpf_cmd cmd, union bpf_attr *attr,
+                             unsigned int size,
+                             struct bpf_common_attr *common_attr,

nit: kernel uses consistent attr_common/size_common pattern, but here
you are inverting attr_common -> common_attr, let's not?

Ack.

I'll keep the same pattern.

quoted

quoted

+                             unsigned int size_common)
+{
+       cmd = common_attr ? (cmd | BPF_COMMON_ATTRS) : (cmd & ~BPF_COMMON_ATTRS);
+       return syscall(__NR_bpf, cmd, attr, size, common_attr, size_common);
+}
+
+static inline int sys_bpf_ext_fd(enum bpf_cmd cmd, union bpf_attr *attr,
+                                unsigned int size,
+                                struct bpf_common_attr *common_attr,
+                                unsigned int size_common)
+{
+       int fd;
+
+       fd = sys_bpf_ext(cmd, attr, size, common_attr, size_common);
+       return ensure_good_fd(fd);
+}
+
+int probe_sys_bpf_ext(void)
+{
+       const size_t attr_sz = offsetofend(union bpf_attr, prog_token_fd);
+       union bpf_attr attr;
+       int fd;
+
+       memset(&attr, 0, attr_sz);
+       fd = syscall(__NR_bpf, BPF_PROG_LOAD | BPF_COMMON_ATTRS, &attr, attr_sz, NULL,
+                    sizeof(struct bpf_common_attr));
+       if (fd >= 0)
+               close(fd);

hm... close can change errno, this is fragile. If fd >= 0, something
is wrong with our detection, just return error right away?

How about capture errno before closing?

err = errno;
if (fd >= 0)
        close(fd);
return err = EFAULT;

not sure what this code is trying to do, but yes, preserving errno is
one way to fix an immediate problem.

But fd should really not be >= 0, and if it is -- it's some problem,
so I'd return an error in that case to keep us aware, which is why I'm
saying I'd just return inside if (fd >= 0) { }

Then, we can wrap all details in probe_sys_bpf_ext().

quoted

quoted

+       return errno == EFAULT;
+}
+

[...]