On Wed Jun 24, 2026 at 4:18 AM PDT, Sun Jian wrote:
> BPF_OBJ_GET_INFO_BY_FD for bpf_link copies struct bpf_link_info back to
> userspace only when ->fill_link_info() succeeds. Some link info providers,
> however, can return -ENOSPC after computing valid metadata when a nested
> userspace output buffer is too small.
>
> For example, perf event tracepoint link info can determine the required
> tp_name length before copying the name fails with -ENOSPC. The current
> top-level error handling returns immediately in that case, so userspace
> observes -ENOSPC but loses the metadata needed to retry with a sufficiently
> large buffer.
>
> Allow bpf_link_get_info_by_fd() to copy the top-level bpf_link_info back
> on -ENOSPC, while still returning -ENOSPC to userspace. Also let perf
> event kprobe, uprobe, and tracepoint link info fill their metadata before
> returning -ENOSPC from nested name buffer copying.
>
> Fixes: f2e10bff16a0 ("bpf: Add support for BPF_OBJ_GET_INFO_BY_FD for
> bpf_link")
> Fixes: 1b715e1b0ec5 ("bpf: Support ->fill_link_info for perf_event")
> Signed-off-by: Sun Jian <[email protected]>
This one is in "wont-fix" category.
Please stop taking AI reports and convert them to pointless patches.
Fix only what humans can hit.
pw-bot: cr
