libbpf: fix BTF dedup to support recursive typedef definitions

Handle recursive typedefs in BTF deduplication

Pahole fails to encode BTF for some Go projects (e.g. Kubernetes and
Podman) due to recursive type definitions that create reference loops
not representable in C. These recursive typedefs trigger a failure in
the BTF deduplication algorithm.

This patch extends btf_dedup_ref_type() to properly handle potential
recursion for BTF_KIND_TYPEDEF, similar to how recursion is already
handled for BTF_KIND_STRUCT. This allows pahole to successfully
generate BTF for Go binaries using recursive types without impacting
existing C-based workflows.

Co-developed-by: Martin Horth <[email protected]>
Signed-off-by: Martin Horth <[email protected]>
Co-developed-by: Ouail Derghal <[email protected]>
Signed-off-by: Ouail Derghal <[email protected]>
Co-developed-by: Guilhem Jazeron <[email protected]>
Signed-off-by: Guilhem Jazeron <[email protected]>
Co-developed-by: Ludovic Paillat <[email protected]>
Signed-off-by: Ludovic Paillat <[email protected]>
Co-developed-by: Robin Theveniaut <[email protected]>
Signed-off-by: Robin Theveniaut <[email protected]>
Suggested-by: Tristan d'Audibert <[email protected]>
Signed-off-by: Paul Houssel <[email protected]>

Author: Paul Houssel

tools/lib/bpf/btf.c

@@ -3901,6 +3901,20 @@ static int btf_dedup_strings(struct btf_dedup *d)
 	return err;
 }
 
+/*
+ * Calculate type signature hash of TYPEDEF, ignoring referenced type IDs,
+ * as referenced type IDs equivalence is established separately during type
+ * graph equivalence check algorithm.
+ */
+static long btf_hash_typedef(struct btf_type *t)
+{
+	long h;
+
+	h = hash_combine(0, t->name_off);
+	h = hash_combine(h, t->info);
+	return h;
+}
+
 static long btf_hash_common(struct btf_type *t)
 {
 	long h;
@@ -3918,6 +3932,13 @@ static bool btf_equal_common(struct btf_type *t1, struct btf_type *t2)
 	       t1->size == t2->size;
 }
 
+/* Check structural compatibility of two TYPEDEF. */
+static bool btf_equal_typedef(struct btf_type *t1, struct btf_type *t2)
+{
+	return t1->name_off == t2->name_off &&
+	       t1->info == t2->info;
+}
+
 /* Calculate type signature hash of INT or TAG. */
 static long btf_hash_int_decl_tag(struct btf_type *t)
 {
@@ -4844,13 +4865,30 @@ static void btf_dedup_merge_hypot_map(struct btf_dedup *d)
 	}
 }
 
+static inline long btf_hash_by_kind(struct btf_type *t, __u16 kind)
+{
+	if (kind == BTF_KIND_TYPEDEF)
+		return btf_hash_typedef(t);
+	else
+		return btf_hash_struct(t);
+}
+
+static inline bool btf_equal_by_kind(struct btf_type *t1, struct btf_type *t2, __u16 kind)
+{
+	if (kind == BTF_KIND_TYPEDEF)
+		return btf_equal_typedef(t1, t2);
+	else
+		return btf_shallow_equal_struct(t1, t2);
+}
+
 /*
- * Deduplicate struct/union types.
+ * Deduplicate struct/union and typedef types.
  *
  * For each struct/union type its type signature hash is calculated, taking
  * into account type's name, size, number, order and names of fields, but
  * ignoring type ID's referenced from fields, because they might not be deduped
- * completely until after reference types deduplication phase. This type hash
+ * completely until after reference types deduplication phase. For each typedef
+ * type, the hash is computed based on the type’s name and size. This type hash
  * is used to iterate over all potential canonical types, sharing same hash.
  * For each canonical candidate we check whether type graphs that they form
  * (through referenced types in fields and so on) are equivalent using algorithm
@@ -4882,26 +4920,28 @@ static int btf_dedup_struct_type(struct btf_dedup *d, __u32 type_id)
 	t = btf_type_by_id(d->btf, type_id);
 	kind = btf_kind(t);
 
-	if (kind != BTF_KIND_STRUCT && kind != BTF_KIND_UNION)
+	if (kind != BTF_KIND_STRUCT &&
+		kind != BTF_KIND_UNION &&
+		kind != BTF_KIND_TYPEDEF)
 		return 0;
 
-	h = btf_hash_struct(t);
+	h = btf_hash_by_kind(t, kind);
 	for_each_dedup_cand(d, hash_entry, h) {
 		__u32 cand_id = hash_entry->value;
 		int eq;
 
 		/*
 		 * Even though btf_dedup_is_equiv() checks for
-		 * btf_shallow_equal_struct() internally when checking two
-		 * structs (unions) for equivalence, we need to guard here
+		 * btf_equal_by_kind() internally when checking two
+		 * structs (unions) or typedefs for equivalence, we need to guard here
 		 * from picking matching FWD type as a dedup candidate.
 		 * This can happen due to hash collision. In such case just
 		 * relying on btf_dedup_is_equiv() would lead to potentially
 		 * creating a loop (FWD -> STRUCT and STRUCT -> FWD), because
 		 * FWD and compatible STRUCT/UNION are considered equivalent.
 		 */
 		cand_type = btf_type_by_id(d->btf, cand_id);
-		if (!btf_shallow_equal_struct(t, cand_type))
+		if (!btf_equal_by_kind(t, cand_type, kind))
 			continue;
 
 		btf_dedup_clear_hypot_map(d);
@@ -4939,18 +4979,18 @@ static int btf_dedup_struct_types(struct btf_dedup *d)
 /*
  * Deduplicate reference type.
  *
- * Once all primitive and struct/union types got deduplicated, we can easily
+ * Once all primitive, struct/union and typedef types got deduplicated, we can easily
  * deduplicate all other (reference) BTF types. This is done in two steps:
  *
  * 1. Resolve all referenced type IDs into their canonical type IDs. This
- * resolution can be done either immediately for primitive or struct/union types
- * (because they were deduped in previous two phases) or recursively for
+ * resolution can be done either immediately for primitive, struct/union, and typedef
+ * types (because they were deduped in previous two phases) or recursively for
  * reference types. Recursion will always terminate at either primitive or
- * struct/union type, at which point we can "unwind" chain of reference types
- * one by one. There is no danger of encountering cycles because in C type
- * system the only way to form type cycle is through struct/union, so any chain
- * of reference types, even those taking part in a type cycle, will inevitably
- * reach struct/union at some point.
+ * struct/union and typedef types, at which point we can "unwind" chain of reference
+ * types one by one. There is no danger of encountering cycles in C, as the only way to
+ * form a type cycle is through struct or union types. Go can form such cycles through
+ * typedef. Thus, any chain of reference types, even those taking part in a type cycle,
+ * will inevitably reach a struct/union or typedef type at some point.
  *
  * 2. Once all referenced type IDs are resolved into canonical ones, BTF type
  * becomes "stable", in the sense that no further deduplication will cause
@@ -4982,7 +5022,6 @@ static int btf_dedup_ref_type(struct btf_dedup *d, __u32 type_id)
 	case BTF_KIND_VOLATILE:
 	case BTF_KIND_RESTRICT:
 	case BTF_KIND_PTR:
-	case BTF_KIND_TYPEDEF:
 	case BTF_KIND_FUNC:
 	case BTF_KIND_TYPE_TAG:
 		ref_type_id = btf_dedup_ref_type(d, t->type);