Re: [PATCH bpf-next] libbpf: Let any two INT/UNION compatible if their names and sizes match

From: Andrii Nakryiko
Date: Tue Nov 30 2021 - 23:17:37 EST


On Tue, Nov 30, 2021 at 7:55 PM Feng zhou <zhoufeng.zf@xxxxxxxxxxxxx> wrote:
>
> From: Feng Zhou <zhoufeng.zf@xxxxxxxxxxxxx>
>
> commit:67c0496e87d193b8356d2af49ab95e8a1b954b3c(kernfs: convert
> kernfs_node->id from union kernfs_node_id to u64).
>
> The bpf program compiles on the kernel version after this commit and
> then tries to run on the kernel before this commit, libbpf will report
> an error. The reverse is also same.
>
> libbpf: prog 'tcp_retransmit_synack_tp': relo #4: kind <byte_off> (0),
> spec is [342] struct kernfs_node.id (0:9 @ offset 104)
> libbpf: prog 'tcp_retransmit_synack_tp': relo #4: non-matching candidate
> libbpf: prog 'tcp_retransmit_synack_tp': relo #4: non-matching candidate
> libbpf: prog 'tcp_retransmit_synack_tp': relo #4: no matching targets
> found
>
> The type before this commit:
> union kernfs_node_id id;
> union kernfs_node_id {
> struct {
> u32 ino;
> u32 generation;
> };
> u64 id;
> };
>
> The type after this commit:
> u64 id;
>
> We can find that the variable name and size have not changed except for
> the type change.
> So I added some judgment to let any two INT/UNION are compatible, if
> their names and sizes match.
>
> Reported-by: Chengming Zhou <zhouchengming@xxxxxxxxxxxxx>
> Tested-by: Chengming Zhou <zhouchengming@xxxxxxxxxxxxx>
> Signed-off-by: Feng Zhou <zhoufeng.zf@xxxxxxxxxxxxx>
> ---

This should be handled by application, not by hacking libbpf's CO-RE
relocation logic. See [0] for how this should be done with existing
BPF CO-RE mechanisms.

[0] https://nakryiko.com/posts/bpf-core-reference-guide/#handling-incompatible-field-and-type-changes

> tools/lib/bpf/relo_core.c | 21 +++++++++++++++++----
> 1 file changed, 17 insertions(+), 4 deletions(-)
>
> diff --git a/tools/lib/bpf/relo_core.c b/tools/lib/bpf/relo_core.c
> index b5b8956a1be8..ff7f4e97bafb 100644
> --- a/tools/lib/bpf/relo_core.c
> +++ b/tools/lib/bpf/relo_core.c
> @@ -294,6 +294,7 @@ static int bpf_core_parse_spec(const struct btf *btf,
> * - any two FLOATs are always compatible;
> * - for ARRAY, dimensionality is ignored, element types are checked for
> * compatibility recursively;
> + * - any two INT/UNION are compatible, if their names and sizes match;
> * - everything else shouldn't be ever a target of relocation.
> * These rules are not set in stone and probably will be adjusted as we get
> * more experience with using BPF CO-RE relocations.
> @@ -313,8 +314,14 @@ static int bpf_core_fields_are_compat(const struct btf *local_btf,
>
> if (btf_is_composite(local_type) && btf_is_composite(targ_type))
> return 1;
> - if (btf_kind(local_type) != btf_kind(targ_type))
> - return 0;
> + if (btf_kind(local_type) != btf_kind(targ_type)) {
> + if (local_type->size == targ_type->size &&
> + (btf_is_union(local_type) || btf_is_union(targ_type)) &&
> + (btf_is_int(local_type) || btf_is_int(targ_type)))
> + return 1;
> + else
> + return 0;
> + }
>
> switch (btf_kind(local_type)) {
> case BTF_KIND_PTR:
> @@ -384,11 +391,17 @@ static int bpf_core_match_member(const struct btf *local_btf,
> targ_type = skip_mods_and_typedefs(targ_btf, targ_id, &targ_id);
> if (!targ_type)
> return -EINVAL;
> - if (!btf_is_composite(targ_type))
> - return 0;
>
> local_id = local_acc->type_id;
> local_type = btf__type_by_id(local_btf, local_id);
> + if (!btf_is_composite(targ_type)) {
> + if (local_type->size == targ_type->size &&
> + btf_is_union(local_type) && btf_is_int(targ_type))
> + return 1;
> + else
> + return 0;
> + }
> +
> local_member = btf_members(local_type) + local_acc->idx;
> local_name = btf__name_by_offset(local_btf, local_member->name_off);
>
> --
> 2.11.0
>