Re: [PATCH v8 bpf-next 07/13] bpf: Add btf_struct_ids_match function

On Wed, Jul 29, 2020 at 9:04 AM Jiri Olsa [off-list ref] wrote:

On Tue, Jul 28, 2020 at 04:35:16PM -0700, Andrii Nakryiko wrote:

SNIP

quoted

quoted

diff --git a/include/linux/bpf.h b/include/linux/bpf.h
index bae557ff2da8..c981e258fed3 100644
--- a/include/linux/bpf.h
+++ b/include/linux/bpf.h

@@ -1306,6 +1306,8 @@ int btf_struct_access(struct bpf_verifier_log *log,
                      const struct btf_type *t, int off, int size,
                      enum bpf_access_type atype,
                      u32 *next_btf_id);
+bool btf_struct_ids_match(struct bpf_verifier_log *log,
+                         int off, u32 id, u32 mid);
 int btf_resolve_helper_id(struct bpf_verifier_log *log,
                          const struct bpf_func_proto *fn, int);

diff --git a/kernel/bpf/btf.c b/kernel/bpf/btf.c
index 1ab5fd5bf992..562d4453fad3 100644
--- a/kernel/bpf/btf.c
+++ b/kernel/bpf/btf.c

@@ -4140,6 +4140,35 @@ int btf_struct_access(struct bpf_verifier_log *log,
        return -EINVAL;
 }

+bool btf_struct_ids_match(struct bpf_verifier_log *log,
+                         int off, u32 id, u32 mid)

just realized that if id == mid and off == 0, btf_struct_ids_match()
will return false. Right now verifier is careful to not call
btf_struct_ids_match in such case, but I wonder if it's better to make
that (common) case also work?

quoted

quoted

+{
+       const struct btf_type *type;
+       u32 nid;
+       int err;
+

mid and nid are terrible names, especially as an input argument name.
mid == need_type_id? nid == cur_type_id or something along those
lines?

'mid' was for matching id, 'nid' for nested id ;-)
need_type_id/cur_type_id sound good

nested I guessed, mid was a mystery to me :))

quoted

quoted

+       do {
+               type = btf_type_by_id(btf_vmlinux, id);
+               if (!type)
+                       return false;
+               err = btf_struct_walk(log, type, off, 1, &nid);
+               if (err < 0)
+                       return false;
+
+               /* We found nested struct object. If it matches
+                * the requested ID, we're done. Otherwise let's
+                * continue the search with offset 0 in the new
+                * type.
+                */
+               if (err == walk_struct && mid == nid)
+                       return true;
+               off = 0;
+               id = nid;
+       } while (err == walk_struct);

This seems like a slightly more obvious control flow:

again:

   ...

   if (err != walk_struct)
      return false;

ok, and perhaps use in here the switch(err) as in the previous patch?

I think straightforward if is better than switch here, because
anything but walk_struct is not what we expect.

thanks,
jirka

quoted

   if (mid != nid) {
      off = 0;
      id = nid;
      goto again;
   }

   return true;

quoted

+
+       return false;
+}
+
 int btf_resolve_helper_id(struct bpf_verifier_log *log,
                          const struct bpf_func_proto *fn, int arg)
 {

[...]