Message-ID: <20200818223921.2911963-6-andriin@fb.com>
Date:   Tue, 18 Aug 2020 15:39:17 -0700
From:   Andrii Nakryiko <andriin@...com>
To:     <bpf@...r.kernel.org>, <netdev@...r.kernel.org>, <ast@...com>,
        <daniel@...earbox.net>
CC:     <andrii.nakryiko@...il.com>, <kernel-team@...com>,
        Andrii Nakryiko <andriin@...com>
Subject: [PATCH bpf-next 5/9] libbpf: implement type-based CO-RE relocations support

Implement support for TYPE_EXISTS/TYPE_SIZE/TYPE_ID_LOCAL/TYPE_ID_REMOTE
relocations. These are examples of type-based relocations, as opposed to
field-based relocations supported already. The difference is that they are
calculating relocation values based on the type itself, not a field within
a struct/union.

Type-based relos have slightly different semantics when matching local types
to kernel target types, see comments in bpf_core_types_are_compat() for
details. Their behavior on failure to find target type in kernel BTF also
differs. Instead of "poisoning" relocatable instruction and failing load
subsequently in kernel, they return 0 (which is rarely a valid return result,
so user BPF code can use that to detect success/failure of the relocation and
deal with it without extra "guarding" relocations). Also, it's always possible
to check existence of the type in target kernel with TYPE_EXISTS relocation,
similarly to a field-based FIELD_EXISTS.

TYPE_ID_LOCAL relocation is a bit special in that it always succeeds (barring
any libbpf/Clang bugs) and resolved to BTF ID using **local** BTF info of BPF
program itself. Tests in subsequent patches demonstrate the usage and
semantics of new relocations.

Signed-off-by: Andrii Nakryiko <andriin@...com>
---
 tools/lib/bpf/bpf_core_read.h   |  52 ++++++-
 tools/lib/bpf/libbpf.c          | 231 ++++++++++++++++++++++++++++----
 tools/lib/bpf/libbpf_internal.h |   4 +
 3 files changed, 263 insertions(+), 24 deletions(-)

diff --git a/tools/lib/bpf/bpf_core_read.h b/tools/lib/bpf/bpf_core_read.h
index 03152cb143b7..684bfb86361a 100644
--- a/tools/lib/bpf/bpf_core_read.h
+++ b/tools/lib/bpf/bpf_core_read.h
@@ -19,6 +19,18 @@ enum bpf_field_info_kind {
 	BPF_FIELD_RSHIFT_U64 = 5,
 };
 
+/* second argument to __builtin_btf_type_id() built-in */
+enum bpf_type_id_kind {
+	BPF_TYPE_ID_LOCAL = 0,		/* BTF type ID in local program */
+	BPF_TYPE_ID_TARGET = 1,		/* BTF type ID in target kernel */
+};
+
+/* second argument to __builtin_preserve_type_info() built-in */
+enum bpf_type_info_kind {
+	BPF_TYPE_EXISTS = 0,		/* type existence in target kernel */
+	BPF_TYPE_SIZE = 1,		/* type size in target kernel */
+};
+
 #define __CORE_RELO(src, field, info)					      \
 	__builtin_preserve_field_info((src)->field, BPF_FIELD_##info)
 
@@ -94,12 +106,50 @@ enum bpf_field_info_kind {
 	__builtin_preserve_field_info(field, BPF_FIELD_EXISTS)
 
 /*
- * Convenience macro to get byte size of a field. Works for integers,
+ * Convenience macro to get the byte size of a field. Works for integers,
  * struct/unions, pointers, arrays, and enums.
  */
 #define bpf_core_field_size(field)					    \
 	__builtin_preserve_field_info(field, BPF_FIELD_BYTE_SIZE)
 
+/*
+ * Convenience macro to get BTF type ID of a specified type, using a local BTF
+ * information. Return 32-bit unsigned integer with type ID from program's own
+ * BTF. Always succeeds.
+ */
+#define bpf_core_type_id_local(type)					    \
+	__builtin_btf_type_id(*(typeof(type) *)0, BPF_TYPE_ID_LOCAL)
+
+/*
+ * Convenience macro to get BTF type ID of a target kernel's type that matches
+ * specified local type.
+ * Returns:
+ *    - valid 32-bit unsigned type ID in kernel BTF;
+ *    - 0, if no matching type was found in a target kernel BTF.
+ */
+#define bpf_core_type_id_kernel(type)					    \
+	__builtin_btf_type_id(*(typeof(type) *)0, BPF_TYPE_ID_TARGET)
+
+/*
+ * Convenience macro to check that provided named type
+ * (struct/union/enum/typedef) exists in a target kernel.
+ * Returns:
+ *    1, if such type is present in target kernel's BTF;
+ *    0, if no matching type is found.
+ */
+#define bpf_core_type_exists(type)					    \
+	__builtin_preserve_type_info(*(typeof(type) *)0, BPF_TYPE_EXISTS)
+
+/*
+ * Convenience macro to get the byte size of a provided named type
+ * (struct/union/enum/typedef) in a target kernel.
+ * Returns:
+ *    >= 0 size (in bytes), if type is present in target kernel's BTF;
+ *    0, if no matching type is found.
+ */
+#define bpf_core_type_size(type)					    \
+	__builtin_preserve_type_info(*(typeof(type) *)0, BPF_TYPE_SIZE)
+
 /*
  * bpf_core_read() abstracts away bpf_probe_read_kernel() call and captures
  * offset relocation for source address using __builtin_preserve_access_index()
diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c
index 1ba458140f50..0379828d3c40 100644
--- a/tools/lib/bpf/libbpf.c
+++ b/tools/lib/bpf/libbpf.c
@@ -4117,6 +4117,10 @@ static const char *core_relo_kind_str(enum bpf_core_relo_kind kind)
 	case BPF_FIELD_SIGNED: return "signed";
 	case BPF_FIELD_LSHIFT_U64: return "lshift_u64";
 	case BPF_FIELD_RSHIFT_U64: return "rshift_u64";
+	case BPF_TYPE_ID_LOCAL: return "local_type_id";
+	case BPF_TYPE_ID_TARGET: return "target_type_id";
+	case BPF_TYPE_EXISTS: return "type_exists";
+	case BPF_TYPE_SIZE: return "type_size";
 	default: return "unknown";
 	}
 }
@@ -4136,6 +4140,19 @@ static bool core_relo_is_field_based(enum bpf_core_relo_kind kind)
 	}
 }
 
+static bool core_relo_is_type_based(enum bpf_core_relo_kind kind)
+{
+	switch (kind) {
+	case BPF_TYPE_ID_LOCAL:
+	case BPF_TYPE_ID_TARGET:
+	case BPF_TYPE_EXISTS:
+	case BPF_TYPE_SIZE:
+		return true;
+	default:
+		return false;
+	}
+}
+
 /*
  * Turn bpf_core_relo into a low- and high-level spec representation,
  * validating correctness along the way, as well as calculating resulting
@@ -4166,6 +4183,9 @@ static bool core_relo_is_field_based(enum bpf_core_relo_kind kind)
  *   - field 'a' access (corresponds to '2' in low-level spec);
  *   - array element #3 access (corresponds to '3' in low-level spec).
  *
+ * Type-based relocations (TYPE_EXISTS/TYPE_SIZE,
+ * TYPE_ID_LOCAL/TYPE_ID_TARGET) don't capture any field information. Their
+ * spec and raw_spec are kept empty.
  */
 static int bpf_core_parse_spec(const struct btf *btf,
 			       __u32 type_id,
@@ -4188,6 +4208,13 @@ static int bpf_core_parse_spec(const struct btf *btf,
 	spec->root_type_id = type_id;
 	spec->relo_kind = relo_kind;
 
+	/* type-based relocations don't have a field access string */
+	if (core_relo_is_type_based(relo_kind)) {
+		if (strcmp(spec_str, "0"))
+			return -EINVAL;
+		return 0;
+	}
+
 	/* parse spec_str="0:1:2:3:4" into array raw_spec=[0, 1, 2, 3, 4] */
 	while (*spec_str) {
 		if (*spec_str == ':')
@@ -4323,7 +4350,7 @@ static struct ids_vec *bpf_core_find_cands(const struct btf *local_btf,
 					   const struct btf *targ_btf)
 {
 	size_t local_essent_len, targ_essent_len;
-	const char *local_name, *targ_name, *targ_kind;
+	const char *local_name, *targ_name;
 	const struct btf_type *t, *local_t;
 	struct ids_vec *cand_ids;
 	__u32 *new_ids;
@@ -4345,13 +4372,11 @@ static struct ids_vec *bpf_core_find_cands(const struct btf *local_btf,
 	n = btf__get_nr_types(targ_btf);
 	for (i = 1; i <= n; i++) {
 		t = btf__type_by_id(targ_btf, i);
-		targ_name = btf__name_by_offset(targ_btf, t->name_off);
-		if (str_is_empty(targ_name))
+		if (btf_kind(t) != btf_kind(local_t))
 			continue;
-		targ_kind = btf_kind_str(t);
 
-		t = skip_mods_and_typedefs(targ_btf, i, NULL);
-		if (!btf_is_composite(t) && !btf_is_array(t))
+		targ_name = btf__name_by_offset(targ_btf, t->name_off);
+		if (str_is_empty(targ_name))
 			continue;
 
 		targ_essent_len = bpf_core_essential_name_len(targ_name);
@@ -4361,7 +4386,7 @@ static struct ids_vec *bpf_core_find_cands(const struct btf *local_btf,
 		if (strncmp(local_name, targ_name, local_essent_len) == 0) {
 			pr_debug("CO-RE relocating [%d] %s %s: found target candidate [%d] %s %s\n",
 				 local_type_id, btf_kind_str(local_t),
-				 local_name, i, targ_kind, targ_name);
+				 local_name, i, btf_kind_str(t), targ_name);
 			new_ids = reallocarray(cand_ids->data,
 					       cand_ids->len + 1,
 					       sizeof(*cand_ids->data));
@@ -4379,8 +4404,9 @@ static struct ids_vec *bpf_core_find_cands(const struct btf *local_btf,
 	return ERR_PTR(err);
 }
 
-/* Check two types for compatibility, skipping const/volatile/restrict and
- * typedefs, to ensure we are relocating compatible entities:
+/* Check two types for compatibility for the purpose of field access
+ * relocation. const/volatile/restrict and typedefs are skipped to ensure we
+ * are relocating semantically compatible entities:
  *   - any two STRUCTs/UNIONs are compatible and can be mixed;
  *   - any two FWDs are compatible, if their names match (modulo flavor suffix);
  *   - any two PTRs are always compatible;
@@ -4535,6 +4561,100 @@ static int bpf_core_match_member(const struct btf *local_btf,
 	return 0;
 }
 
+/* Check local and target types for compatibility. This check is used for
+ * type-based CO-RE relocations and follow slightly different rules than
+ * field-based relocations. This function assumes that root types were already
+ * checked for name match. Beyond that initial root-level name check, names
+ * are completely ignored. Compatibility rules are as follows:
+ *   - any two STRUCTs/UNIONs/FWDs/ENUMs/INTs are considered compatible, but
+ *     kind should match for local and target types (i.e., STRUCT is not
+ *     compatible with UNION);
+ *   - for ENUMs, the size is ignored;
+ *   - for INT, size and signedness are ignored;
+ *   - for ARRAY, dimensionality is ignored, element types are checked for
+ *     compatibility recursively;
+ *   - CONST/VOLATILE/RESTRICT modifiers are ignored;
+ *   - TYPEDEFs/PTRs are compatible if types they pointing to are compatible;
+ *   - FUNC_PROTOs are compatible if they have compatible signature: same
+ *     number of input args and compatible return and argument types.
+ * These rules are not set in stone and probably will be adjusted as we get
+ * more experience with using BPF CO-RE relocations.
+ */
+static int bpf_core_types_are_compat(const struct btf *local_btf, __u32 local_id,
+				     const struct btf *targ_btf, __u32 targ_id)
+{
+	const struct btf_type *local_type, *targ_type;
+	int depth = 32; /* max recursion depth */
+
+	/* caller made sure that names match (ignoring flavor suffix) */
+	local_type = btf__type_by_id(local_btf, local_id);
+	targ_type = btf__type_by_id(local_btf, local_id);
+	if (btf_kind(local_type) != btf_kind(targ_type))
+		return 0;
+
+recur:
+	depth--;
+	if (depth < 0)
+		return -EINVAL;
+
+	local_type = skip_mods_and_typedefs(local_btf, local_id, &local_id);
+	targ_type = skip_mods_and_typedefs(targ_btf, targ_id, &targ_id);
+	if (!local_type || !targ_type)
+		return -EINVAL;
+
+	if (btf_kind(local_type) != btf_kind(targ_type))
+		return 0;
+
+	switch (btf_kind(local_type)) {
+	case BTF_KIND_UNKN:
+	case BTF_KIND_STRUCT:
+	case BTF_KIND_UNION:
+	case BTF_KIND_ENUM:
+	case BTF_KIND_FWD:
+		return 1;
+	case BTF_KIND_INT:
+		/* just reject deprecated bitfield-like integers; all other
+		 * integers are by default compatible between each other
+		 */
+		return btf_int_offset(local_type) == 0 && btf_int_offset(targ_type) == 0;
+	case BTF_KIND_PTR:
+		local_id = local_type->type;
+		targ_id = targ_type->type;
+		goto recur;
+	case BTF_KIND_ARRAY:
+		local_id = btf_array(local_type)->type;
+		targ_id = btf_array(targ_type)->type;
+		goto recur;
+	case BTF_KIND_FUNC_PROTO: {
+		struct btf_param *local_p = btf_params(local_type);
+		struct btf_param *targ_p = btf_params(targ_type);
+		__u16 local_vlen = btf_vlen(local_type);
+		__u16 targ_vlen = btf_vlen(targ_type);
+		int i, err;
+
+		if (local_vlen != targ_vlen)
+			return 0;
+
+		for (i = 0; i < local_vlen; i++, local_p++, targ_p++) {
+			skip_mods_and_typedefs(local_btf, local_p->type, &local_id);
+			skip_mods_and_typedefs(targ_btf, targ_p->type, &targ_id);
+			err = bpf_core_types_are_compat(local_btf, local_id, targ_btf, targ_id);
+			if (err <= 0)
+				return err;
+		}
+
+		/* tail recurse for return type check */
+		skip_mods_and_typedefs(local_btf, local_type->type, &local_id);
+		skip_mods_and_typedefs(targ_btf, targ_type->type, &targ_id);
+		goto recur;
+	}
+	default:
+		pr_warn("unexpected kind %s relocated, local [%d], target [%d]\n",
+			btf_kind_str(local_type), local_id, targ_id);
+		return 0;
+	}
+}
+
 /*
  * Try to match local spec to a target type and, if successful, produce full
  * target spec (high-level, low-level + bit offset).
@@ -4553,6 +4673,12 @@ static int bpf_core_spec_match(struct bpf_core_spec *local_spec,
 	targ_spec->root_type_id = targ_id;
 	targ_spec->relo_kind = local_spec->relo_kind;
 
+	if (core_relo_is_type_based(local_spec->relo_kind)) {
+		return bpf_core_types_are_compat(local_spec->btf,
+						 local_spec->root_type_id,
+						 targ_btf, targ_id);
+	}
+
 	local_acc = &local_spec->spec[0];
 	targ_acc = &targ_spec->spec[0];
 
@@ -4726,6 +4852,40 @@ static int bpf_core_calc_field_relo(const struct bpf_program *prog,
 	return 0;
 }
 
+static int bpf_core_calc_type_relo(const struct bpf_core_relo *relo,
+				   const struct bpf_core_spec *spec,
+				   __u32 *val)
+{
+	__s64 sz;
+
+	/* type-based relos return zero when target type is not found */
+	if (!spec) {
+		*val = 0;
+		return 0;
+	}
+
+	switch (relo->kind) {
+	case BPF_TYPE_ID_TARGET:
+		*val = spec->root_type_id;
+		break;
+	case BPF_TYPE_EXISTS:
+		*val = 1;
+		break;
+	case BPF_TYPE_SIZE:
+		sz = btf__resolve_size(spec->btf, spec->root_type_id);
+		if (sz < 0)
+			return -EINVAL;
+		*val = sz;
+		break;
+	case BPF_TYPE_ID_LOCAL:
+	/* BPF_TYPE_ID_LOCAL is handled specially and shouldn't get here */
+	default:
+		return -EOPNOTSUPP;
+	}
+
+	return 0;
+}
+
 struct bpf_core_relo_res
 {
 	/* expected value in the instruction, unless validate == false */
@@ -4761,6 +4921,9 @@ static int bpf_core_calc_relo(const struct bpf_program *prog,
 	if (core_relo_is_field_based(relo->kind)) {
 		err = bpf_core_calc_field_relo(prog, relo, local_spec, &res->orig_val, &res->validate);
 		err = err ?: bpf_core_calc_field_relo(prog, relo, targ_spec, &res->new_val, NULL);
+	} else if (core_relo_is_type_based(relo->kind)) {
+		err = bpf_core_calc_type_relo(relo, local_spec, &res->orig_val);
+		err = err ?: bpf_core_calc_type_relo(relo, targ_spec, &res->new_val);
 	}
 
 	if (err == -EUCLEAN) {
@@ -4900,6 +5063,9 @@ static void bpf_core_dump_spec(int level, const struct bpf_core_spec *spec)
 
 	libbpf_print(level, "[%u] %s %s", type_id, btf_kind_str(t), str_is_empty(s) ? "<anon>" : s);
 
+	if (core_relo_is_type_based(spec->relo_kind))
+		return;
+
 	if (core_relo_is_field_based(spec->relo_kind)) {
 		for (i = 0; i < spec->len; i++) {
 			if (spec->spec[i].name)
@@ -4917,6 +5083,7 @@ static void bpf_core_dump_spec(int level, const struct bpf_core_spec *spec)
 				     spec->bit_offset / 8, spec->bit_offset % 8);
 		else
 			libbpf_print(level, " @ offset %u)", spec->bit_offset / 8);
+		return;
 	}
 }
 
@@ -4985,12 +5152,12 @@ static void *u32_as_hash_key(__u32 x)
  *    between multiple relocations for the same type ID and is updated as some
  *    of the candidates are pruned due to structural incompatibility.
  */
-static int bpf_core_reloc_field(struct bpf_program *prog,
-				 const struct bpf_core_relo *relo,
-				 int relo_idx,
-				 const struct btf *local_btf,
-				 const struct btf *targ_btf,
-				 struct hashmap *cand_cache)
+static int bpf_core_apply_relo(struct bpf_program *prog,
+			       const struct bpf_core_relo *relo,
+			       int relo_idx,
+			       const struct btf *local_btf,
+			       const struct btf *targ_btf,
+			       struct hashmap *cand_cache)
 {
 	const char *prog_name = bpf_program__title(prog, false);
 	struct bpf_core_spec local_spec, cand_spec, targ_spec;
@@ -5009,7 +5176,7 @@ static int bpf_core_reloc_field(struct bpf_program *prog,
 		return -EINVAL;
 
 	local_name = btf__name_by_offset(local_btf, local_type->name_off);
-	if (str_is_empty(local_name))
+	if (!local_name)
 		return -EINVAL;
 
 	spec_str = btf__name_by_offset(local_btf, relo->access_str_off);
@@ -5020,7 +5187,8 @@ static int bpf_core_reloc_field(struct bpf_program *prog,
 	if (err) {
 		pr_warn("prog '%s': relo #%d: parsing [%d] %s %s + %s failed: %d\n",
 			prog_name, relo_idx, local_id, btf_kind_str(local_type),
-			local_name, spec_str, err);
+			str_is_empty(local_name) ? "<anon>" : local_name,
+			spec_str, err);
 		return -EINVAL;
 	}
 
@@ -5029,12 +5197,28 @@ static int bpf_core_reloc_field(struct bpf_program *prog,
 	bpf_core_dump_spec(LIBBPF_DEBUG, &local_spec);
 	libbpf_print(LIBBPF_DEBUG, "\n");
 
+	/* TYPE_ID_LOCAL relo is special and doesn't need candidate search */
+	if (relo->kind == BPF_TYPE_ID_LOCAL) {
+		targ_res.validate = true;
+		targ_res.poison = false;
+		targ_res.orig_val = local_spec.root_type_id;
+		targ_res.new_val = local_spec.root_type_id;
+		goto patch_insn;
+	}
+
+	/* libbpf doesn't support candidate search for anonymous types */
+	if (str_is_empty(spec_str)) {
+		pr_warn("prog '%s': relo #%d: <%s> (%d) relocation doesn't support anonymous types\n",
+			prog_name, relo_idx, core_relo_kind_str(relo->kind), relo->kind);
+		return -EOPNOTSUPP;
+	}
+
 	if (!hashmap__find(cand_cache, type_key, (void **)&cand_ids)) {
 		cand_ids = bpf_core_find_cands(local_btf, local_id, targ_btf);
 		if (IS_ERR(cand_ids)) {
 			pr_warn("prog '%s': relo #%d: target candidate search failed for [%d] %s %s: %ld",
-				prog_name, relo_idx, local_id, btf_kind_str(local_type), local_name,
-				PTR_ERR(cand_ids));
+				prog_name, relo_idx, local_id, btf_kind_str(local_type),
+				local_name, PTR_ERR(cand_ids));
 			return PTR_ERR(cand_ids);
 		}
 		err = hashmap__set(cand_cache, type_key, cand_ids, NULL, NULL);
@@ -5090,7 +5274,7 @@ static int bpf_core_reloc_field(struct bpf_program *prog,
 			return -EINVAL;
 		}
 
-		cand_ids->data[j++] = cand_spec.spec[0].type_id;
+		cand_ids->data[j++] = cand_spec.root_type_id;
 	}
 
 	/*
@@ -5109,7 +5293,7 @@ static int bpf_core_reloc_field(struct bpf_program *prog,
 	 * as well as expected case, depending whether instruction w/
 	 * relocation is guarded in some way that makes it unreachable (dead
 	 * code) if relocation can't be resolved. This is handled in
-	 * bpf_core_reloc_insn() uniformly by replacing that instruction with
+	 * bpf_core_patch_insn() uniformly by replacing that instruction with
 	 * BPF helper call insn (using invalid helper ID). If that instruction
 	 * is indeed unreachable, then it will be ignored and eliminated by
 	 * verifier. If it was an error, then verifier will complain and point
@@ -5125,6 +5309,7 @@ static int bpf_core_reloc_field(struct bpf_program *prog,
 			return err;
 	}
 
+patch_insn:
 	/* bpf_core_patch_insn() should know how to handle missing targ_spec */
 	err = bpf_core_patch_insn(prog, relo, relo_idx, &targ_res);
 	if (err) {
@@ -5192,8 +5377,8 @@ bpf_object__relocate_core(struct bpf_object *obj, const char *targ_btf_path)
 			 sec_name, sec->num_info);
 
 		for_each_btf_ext_rec(seg, sec, i, rec) {
-			err = bpf_core_reloc_field(prog, rec, i, obj->btf,
-						   targ_btf, cand_cache);
+			err = bpf_core_apply_relo(prog, rec, i, obj->btf,
+						  targ_btf, cand_cache);
 			if (err) {
 				pr_warn("prog '%s': relo #%d: failed to relocate: %d\n",
 					sec_name, i, err);
diff --git a/tools/lib/bpf/libbpf_internal.h b/tools/lib/bpf/libbpf_internal.h
index b776a7125c92..30eb2cf33789 100644
--- a/tools/lib/bpf/libbpf_internal.h
+++ b/tools/lib/bpf/libbpf_internal.h
@@ -218,6 +218,10 @@ enum bpf_core_relo_kind {
 	BPF_FIELD_SIGNED = 3,		/* field signedness (0 - unsigned, 1 - signed) */
 	BPF_FIELD_LSHIFT_U64 = 4,	/* bitfield-specific left bitshift */
 	BPF_FIELD_RSHIFT_U64 = 5,	/* bitfield-specific right bitshift */
+	BPF_TYPE_ID_LOCAL = 6,		/* type ID in local BPF object */
+	BPF_TYPE_ID_TARGET = 7,		/* type ID in target kernel */
+	BPF_TYPE_EXISTS = 8,		/* type existence in target kernel */
+	BPF_TYPE_SIZE = 9,		/* type size in bytes */
 };
 
 /* The minimum bpf_core_relo checked by the loader
-- 
2.24.1

Powered by blists - more mailing lists