On Fri, Nov 19, 2021 at 7:33 PM Alexei Starovoitov
[off-list ref] wrote:

From: Alexei Starovoitov <ast@kernel.org>

Given BPF program's BTF perform a linear search through kernel BTFs for
a possible candidate.
Then wire the result into bpf_core_apply_relo_insn().

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
---
 kernel/bpf/btf.c | 136 ++++++++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 135 insertions(+), 1 deletion(-)

[...]

 int bpf_core_apply(struct bpf_core_ctx *ctx, const struct bpf_core_relo *relo,
                   int relo_idx, void *insn)
 {
-       return -EOPNOTSUPP;
+       if (relo->kind != BPF_CORE_TYPE_ID_LOCAL) {
+               struct bpf_core_cand_list *cands;
+
+               cands = bpf_core_find_cands(ctx, relo->type_id);

this is wrong for many reasons:

1. you will overwrite previous ctx->cands, if it was already set,
which leaks memory
2. this list of candidates should be keyed by relo->type_id ("root
type"). Different root types get their own independent lists; so it
has to be some sort of look up table from type_id to a list of
candidates.

2) means that if you had a bunch of relos against struct task_struct,
you'll crate a list of candidates when processing first relo that
starts at task_struct. All the subsequent relos that have task_struct
as root type will re-used that list and potentially trim it down. If
there are some other relos against, say, struct mm_struct, they will
have their independent list of candidates.

+               if (IS_ERR(cands)) {
+                       bpf_log(ctx->log, "target candidate search failed for %d\n",
+                               relo->type_id);
+                       return PTR_ERR(cands);
+               }
+               ctx->cands = cands;
+       }
+       return bpf_core_apply_relo_insn((void *)ctx->log, insn, relo->insn_off / 8,
+                                       relo, relo_idx, ctx->btf, ctx->cands);
 }
--
2.30.2