1 files changed, 105 insertions, 34 deletions

diff --git a/tools/lib/bpf/btf.c b/tools/lib/bpf/btf.c
index fadf089ae8fe..9aa19c89f758 100644
--- a/tools/lib/bpf/btf.c
+++ b/tools/lib/bpf/btf.c
@@ -454,7 +454,7 @@ const struct btf *btf__base_btf(const struct btf *btf)
 }
 
 /* internal helper returning non-const pointer to a type */
-struct btf_type *btf_type_by_id(struct btf *btf, __u32 type_id)
+struct btf_type *btf_type_by_id(const struct btf *btf, __u32 type_id)
 {
 	if (type_id == 0)
 		return &btf_void;
@@ -610,6 +610,7 @@ __s64 btf__resolve_size(const struct btf *btf, __u32 type_id)
 		case BTF_KIND_RESTRICT:
 		case BTF_KIND_VAR:
 		case BTF_KIND_DECL_TAG:
+		case BTF_KIND_TYPE_TAG:
 			type_id = t->type;
 			break;
 		case BTF_KIND_ARRAY:
@@ -1123,54 +1124,86 @@ struct btf *btf__parse_split(const char *path, struct btf *base_btf)
 
 static void *btf_get_raw_data(const struct btf *btf, __u32 *size, bool swap_endian);
 
-int btf__load_into_kernel(struct btf *btf)
+int btf_load_into_kernel(struct btf *btf, char *log_buf, size_t log_sz, __u32 log_level)
 {
-	__u32 log_buf_size = 0, raw_size;
-	char *log_buf = NULL;
+	LIBBPF_OPTS(bpf_btf_load_opts, opts);
+	__u32 buf_sz = 0, raw_size;
+	char *buf = NULL, *tmp;
 	void *raw_data;
 	int err = 0;
 
 	if (btf->fd >= 0)
 		return libbpf_err(-EEXIST);
+	if (log_sz && !log_buf)
+		return libbpf_err(-EINVAL);
 
-retry_load:
-	if (log_buf_size) {
-		log_buf = malloc(log_buf_size);
-		if (!log_buf)
-			return libbpf_err(-ENOMEM);
-
-		*log_buf = 0;
-	}
-
+	/* cache native raw data representation */
 	raw_data = btf_get_raw_data(btf, &raw_size, false);
 	if (!raw_data) {
 		err = -ENOMEM;
 		goto done;
 	}
-	/* cache native raw data representation */
 	btf->raw_size = raw_size;
 	btf->raw_data = raw_data;
 
-	btf->fd = bpf_load_btf(raw_data, raw_size, log_buf, log_buf_size, false);
+retry_load:
+	/* if log_level is 0, we won't provide log_buf/log_size to the kernel,
+	 * initially. Only if BTF loading fails, we bump log_level to 1 and
+	 * retry, using either auto-allocated or custom log_buf. This way
+	 * non-NULL custom log_buf provides a buffer just in case, but hopes
+	 * for successful load and no need for log_buf.
+	 */
+	if (log_level) {
+		/* if caller didn't provide custom log_buf, we'll keep
+		 * allocating our own progressively bigger buffers for BTF
+		 * verification log
+		 */
+		if (!log_buf) {
+			buf_sz = max((__u32)BPF_LOG_BUF_SIZE, buf_sz * 2);
+			tmp = realloc(buf, buf_sz);
+			if (!tmp) {
+				err = -ENOMEM;
+				goto done;
+			}
+			buf = tmp;
+			buf[0] = '\0';
+		}
+
+		opts.log_buf = log_buf ? log_buf : buf;
+		opts.log_size = log_buf ? log_sz : buf_sz;
+		opts.log_level = log_level;
+	}
+
+	btf->fd = bpf_btf_load(raw_data, raw_size, &opts);
 	if (btf->fd < 0) {
-		if (!log_buf || errno == ENOSPC) {
-			log_buf_size = max((__u32)BPF_LOG_BUF_SIZE,
-					   log_buf_size << 1);
-			free(log_buf);
+		/* time to turn on verbose mode and try again */
+		if (log_level == 0) {
+			log_level = 1;
 			goto retry_load;
 		}
+		/* only retry if caller didn't provide custom log_buf, but
+		 * make sure we can never overflow buf_sz
+		 */
+		if (!log_buf && errno == ENOSPC && buf_sz <= UINT_MAX / 2)
+			goto retry_load;
 
 		err = -errno;
-		pr_warn("Error loading BTF: %s(%d)\n", strerror(errno), errno);
-		if (*log_buf)
-			pr_warn("%s\n", log_buf);
-		goto done;
+		pr_warn("BTF loading error: %d\n", err);
+		/* don't print out contents of custom log_buf */
+		if (!log_buf && buf[0])
+			pr_warn("-- BEGIN BTF LOAD LOG ---\n%s\n-- END BTF LOAD LOG --\n", buf);
 	}
 
 done:
-	free(log_buf);
+	free(buf);
 	return libbpf_err(err);
 }
+
+int btf__load_into_kernel(struct btf *btf)
+{
+	return btf_load_into_kernel(btf, NULL, 0, 0);
+}
+
 int btf__load(struct btf *) __attribute__((alias("btf__load_into_kernel")));
 
 int btf__fd(const struct btf *btf)
@@ -2730,15 +2763,11 @@ void btf_ext__free(struct btf_ext *btf_ext)
 	free(btf_ext);
 }
 
-struct btf_ext *btf_ext__new(__u8 *data, __u32 size)
+struct btf_ext *btf_ext__new(const __u8 *data, __u32 size)
 {
 	struct btf_ext *btf_ext;
 	int err;
 
-	err = btf_ext_parse_hdr(data, size);
-	if (err)
-		return libbpf_err_ptr(err);
-
 	btf_ext = calloc(1, sizeof(struct btf_ext));
 	if (!btf_ext)
 		return libbpf_err_ptr(-ENOMEM);
@@ -2751,6 +2780,10 @@ struct btf_ext *btf_ext__new(__u8 *data, __u32 size)
 	}
 	memcpy(btf_ext->data, data, size);
 
+	err = btf_ext_parse_hdr(btf_ext->data, size);
+	if (err)
+		goto done;
+
 	if (btf_ext->hdr->hdr_len < offsetofend(struct btf_ext_header, line_info_len)) {
 		err = -EINVAL;
 		goto done;
@@ -3074,7 +3107,7 @@ done:
 	return libbpf_err(err);
 }
 
-COMPAT_VERSION(bpf__dedup_deprecated, btf__dedup, LIBBPF_0.0.2)
+COMPAT_VERSION(btf__dedup_deprecated, btf__dedup, LIBBPF_0.0.2)
 int btf__dedup_deprecated(struct btf *btf, struct btf_ext *btf_ext, const void *unused_opts)
 {
 	LIBBPF_OPTS(btf_dedup_opts, opts, .btf_ext = btf_ext);
@@ -3476,8 +3509,8 @@ static long btf_hash_struct(struct btf_type *t)
 }
 
 /*
- * Check structural compatibility of two FUNC_PROTOs, ignoring referenced type
- * IDs. This check is performed during type graph equivalence check and
+ * Check structural compatibility of two STRUCTs/UNIONs, ignoring referenced
+ * type IDs. This check is performed during type graph equivalence check and
  * referenced types equivalence is checked separately.
  */
 static bool btf_shallow_equal_struct(struct btf_type *t1, struct btf_type *t2)
@@ -3850,6 +3883,31 @@ static int btf_dedup_identical_arrays(struct btf_dedup *d, __u32 id1, __u32 id2)
 	return btf_equal_array(t1, t2);
 }
 
+/* Check if given two types are identical STRUCT/UNION definitions */
+static bool btf_dedup_identical_structs(struct btf_dedup *d, __u32 id1, __u32 id2)
+{
+	const struct btf_member *m1, *m2;
+	struct btf_type *t1, *t2;
+	int n, i;
+
+	t1 = btf_type_by_id(d->btf, id1);
+	t2 = btf_type_by_id(d->btf, id2);
+
+	if (!btf_is_composite(t1) || btf_kind(t1) != btf_kind(t2))
+		return false;
+
+	if (!btf_shallow_equal_struct(t1, t2))
+		return false;
+
+	m1 = btf_members(t1);
+	m2 = btf_members(t2);
+	for (i = 0, n = btf_vlen(t1); i < n; i++, m1++, m2++) {
+		if (m1->type != m2->type)
+			return false;
+	}
+	return true;
+}
+
 /*
  * Check equivalence of BTF type graph formed by candidate struct/union (we'll
  * call it "candidate graph" in this description for brevity) to a type graph
@@ -3961,6 +4019,8 @@ static int btf_dedup_is_equiv(struct btf_dedup *d, __u32 cand_id,
 
 	hypot_type_id = d->hypot_map[canon_id];
 	if (hypot_type_id <= BTF_MAX_NR_TYPES) {
+		if (hypot_type_id == cand_id)
+			return 1;
 		/* In some cases compiler will generate different DWARF types
 		 * for *identical* array type definitions and use them for
 		 * different fields within the *same* struct. This breaks type
@@ -3969,8 +4029,18 @@ static int btf_dedup_is_equiv(struct btf_dedup *d, __u32 cand_id,
 		 * types within a single CU. So work around that by explicitly
 		 * allowing identical array types here.
 		 */
-		return hypot_type_id == cand_id ||
-		       btf_dedup_identical_arrays(d, hypot_type_id, cand_id);
+		if (btf_dedup_identical_arrays(d, hypot_type_id, cand_id))
+			return 1;
+		/* It turns out that similar situation can happen with
+		 * struct/union sometimes, sigh... Handle the case where
+		 * structs/unions are exactly the same, down to the referenced
+		 * type IDs. Anything more complicated (e.g., if referenced
+		 * types are different, but equivalent) is *way more*
+		 * complicated and requires a many-to-many equivalence mapping.
+		 */
+		if (btf_dedup_identical_structs(d, hypot_type_id, cand_id))
+			return 1;
+		return 0;
 	}
 
 	if (btf_dedup_hypot_map_add(d, canon_id, cand_id))
@@ -4023,6 +4093,7 @@ static int btf_dedup_is_equiv(struct btf_dedup *d, __u32 cand_id,
 	case BTF_KIND_PTR:
 	case BTF_KIND_TYPEDEF:
 	case BTF_KIND_FUNC:
+	case BTF_KIND_TYPE_TAG:
 		if (cand_type->info != canon_type->info)
 			return 0;
 		return btf_dedup_is_equiv(d, cand_type->type, canon_type->type);