Merge branch 'add-fd_array_cnt-attribute-for-bpf_prog_load'

Anton Protopopov says:

====================
Add fd_array_cnt attribute for BPF_PROG_LOAD

Add a new attribute to the bpf(BPF_PROG_LOAD) system call. If this
new attribute is non-zero, then the fd_array is considered to be a
continuous array of the fd_array_cnt length and to contain only
proper map file descriptors or btf file descriptors.

This change allows maps (and btfs), which aren't referenced directly
by a BPF program, to be bound to the program _and_ also to be present
during the program verification (so BPF_PROG_BIND_MAP is not enough
for this use case).

The primary reason for this change is that it is a prerequisite for
adding "instruction set" maps, which are both non-referenced by the
program and must be present during the program verification.

The first five commits add the new functionality, the sixth adds
corresponding self-tests, and the last one is a small additional fix.

v1 -> v2:
* rewrite the add_fd_from_fd_array() function (Eduard)
* a few cleanups in selftests (Eduard)

v2 -> v3:
* various renamings (Alexei)
* "0 is not special" (Alexei, Andrii)
* do not alloc memory on fd_array init (Alexei)
* fix leaking maps for error path (Hou Tao)
* use libbpf helpers vs. raw syscalls (Andrii)
* add comments on __btf_get_by_fd/__bpf_map_get (Alexei)
* remove extra code (Alexei)

v3 -> v4:
* simplify error path when parsing fd_array
* libbpf: pass fd_array_cnt only in prog_load (Alexei)
* selftests patch contained extra code (Alexei)
* renames, fix comments (Alexei)

v4 -> v5:
* Add btfs to env->used_btfs (Andrii)
* Fix an integer overflow (Andrii)
* A set of cleanups for selftests (Andrii)
====================

Link: https://patch.msgid.link/[e-mail]
Signed-off-by: Andrii Nakryiko <[e-mail]>

include/linux/bpf.h

···
 struct bpf_map *bpf_map_get(u32 ufd);
 struct bpf_map *bpf_map_get_with_uref(u32 ufd);
 
+/*
+ * The __bpf_map_get() and __btf_get_by_fd() functions parse a file
+ * descriptor and return a corresponding map or btf object.
+ * Their names are double underscored to emphasize the fact that they
+ * do not increase refcnt. To also increase refcnt use corresponding
+ * bpf_map_get() and btf_get_by_fd() functions.
+ */
+
 static inline struct bpf_map *__bpf_map_get(struct fd f)
 {
 	if (fd_empty(f))
 		return ERR_PTR(-EBADF);
 	if (unlikely(fd_file(f)->f_op != &bpf_map_fops))
+		return ERR_PTR(-EINVAL);
+	return fd_file(f)->private_data;
+}
+
+static inline struct btf *__btf_get_by_fd(struct fd f)
+{
+	if (fd_empty(f))
+		return ERR_PTR(-EBADF);
+	if (unlikely(fd_file(f)->f_op != &btf_fops))
 		return ERR_PTR(-EINVAL);
 	return fd_file(f)->private_data;
 }

include/uapi/linux/bpf.h

···
 		 * If provided, prog_flags should have BPF_F_TOKEN_FD flag set.
 		 */
 		__s32		prog_token_fd;
+		/* The fd_array_cnt can be used to pass the length of the
+		 * fd_array array. In this case all the [map] file descriptors
+		 * passed in this array will be bound to the program, even if
+		 * the maps are not referenced directly. The functionality is
+		 * similar to the BPF_PROG_BIND_MAP syscall, but maps can be
+		 * used by the verifier during the program load. If provided,
+		 * then the fd_array[0,...,fd_array_cnt-1] is expected to be
+		 * continuous.
+		 */
+		__u32		fd_array_cnt;
 	};
 
 	struct { /* anonymous struct used by BPF_OBJ_* commands */

kernel/bpf/btf.c

···
 	struct btf *btf;
 	CLASS(fd, f)(fd);
 
-	if (fd_empty(f))
-		return ERR_PTR(-EBADF);
-
-	if (fd_file(f)->f_op != &btf_fops)
-		return ERR_PTR(-EINVAL);
-
-	btf = fd_file(f)->private_data;
-	refcount_inc(&btf->refcnt);
+	btf = __btf_get_by_fd(f);
+	if (!IS_ERR(btf))
+		refcount_inc(&btf->refcnt);
 
 	return btf;
 }

kernel/bpf/syscall.c

···
 }
 
 /* last field in 'union bpf_attr' used by this command */
-#define BPF_PROG_LOAD_LAST_FIELD prog_token_fd
+#define BPF_PROG_LOAD_LAST_FIELD fd_array_cnt
 
 static int bpf_prog_load(union bpf_attr *attr, bpfptr_t uattr, u32 uattr_size)
 {

kernel/bpf/verifier.c

···
 	return -ENOENT;
 }
 
+/*
+ * Add btf to the used_btfs array and return the index. (If the btf was
+ * already added, then just return the index.) Upon successful insertion
+ * increase btf refcnt, and, if present, also refcount the corresponding
+ * kernel module.
+ */
+static int __add_used_btf(struct bpf_verifier_env *env, struct btf *btf)
+{
+	struct btf_mod_pair *btf_mod;
+	int i;
+
+	/* check whether we recorded this BTF (and maybe module) already */
+	for (i = 0; i < env->used_btf_cnt; i++)
+		if (env->used_btfs[i].btf == btf)
+			return i;
+
+	if (env->used_btf_cnt >= MAX_USED_BTFS)
+		return -E2BIG;
+
+	btf_get(btf);
+
+	btf_mod = &env->used_btfs[env->used_btf_cnt];
+	btf_mod->btf = btf;
+	btf_mod->module = NULL;
+
+	/* if we reference variables from kernel module, bump its refcount */
+	if (btf_is_module(btf)) {
+		btf_mod->module = btf_try_get_module(btf);
+		if (!btf_mod->module) {
+			btf_put(btf);
+			return -ENXIO;
+		}
+	}
+
+	return env->used_btf_cnt++;
+}
+
 /* replace pseudo btf_id with kernel symbol address */
-static int check_pseudo_btf_id(struct bpf_verifier_env *env,
-			       struct bpf_insn *insn,
-			       struct bpf_insn_aux_data *aux)
+static int __check_pseudo_btf_id(struct bpf_verifier_env *env,
+				 struct bpf_insn *insn,
+				 struct bpf_insn_aux_data *aux,
+				 struct btf *btf)
 {
 	const struct btf_var_secinfo *vsi;
 	const struct btf_type *datasec;
-	struct btf_mod_pair *btf_mod;
 	const struct btf_type *t;
 	const char *sym_name;
 	bool percpu = false;
 	u32 type, id = insn->imm;
-	struct btf *btf;
 	s32 datasec_id;
 	u64 addr;
-	int i, btf_fd, err;
-
-	btf_fd = insn[1].imm;
-	if (btf_fd) {
-		btf = btf_get_by_fd(btf_fd);
-		if (IS_ERR(btf)) {
-			verbose(env, "invalid module BTF object FD specified.\n");
-			return -EINVAL;
-		}
-	} else {
-		if (!btf_vmlinux) {
-			verbose(env, "kernel is missing BTF, make sure CONFIG_DEBUG_INFO_BTF=y is specified in Kconfig.\n");
-			return -EINVAL;
-		}
-		btf = btf_vmlinux;
-		btf_get(btf);
-	}
+	int i;
 
 	t = btf_type_by_id(btf, id);
 	if (!t) {
 		verbose(env, "ldimm64 insn specifies invalid btf_id %d.\n", id);
-		err = -ENOENT;
-		goto err_put;
+		return -ENOENT;
 	}
 
 	if (!btf_type_is_var(t) && !btf_type_is_func(t)) {
 		verbose(env, "pseudo btf_id %d in ldimm64 isn't KIND_VAR or KIND_FUNC\n", id);
-		err = -EINVAL;
-		goto err_put;
+		return -EINVAL;
 	}
 
 	sym_name = btf_name_by_offset(btf, t->name_off);
···
 	if (!addr) {
 		verbose(env, "ldimm64 failed to find the address for kernel symbol '%s'.\n",
 			sym_name);
-		err = -ENOENT;
-		goto err_put;
+		return -ENOENT;
 	}
 	insn[0].imm = (u32)addr;
 	insn[1].imm = addr >> 32;
···
 	if (btf_type_is_func(t)) {
 		aux->btf_var.reg_type = PTR_TO_MEM | MEM_RDONLY;
 		aux->btf_var.mem_size = 0;
-		goto check_btf;
+		return 0;
 	}
 
 	datasec_id = find_btf_percpu_datasec(btf);
···
 			tname = btf_name_by_offset(btf, t->name_off);
 			verbose(env, "ldimm64 unable to resolve the size of type '%s': %ld\n",
 				tname, PTR_ERR(ret));
-			err = -EINVAL;
-			goto err_put;
+			return -EINVAL;
 		}
 		aux->btf_var.reg_type = PTR_TO_MEM | MEM_RDONLY;
 		aux->btf_var.mem_size = tsize;
···
 		aux->btf_var.btf = btf;
 		aux->btf_var.btf_id = type;
 	}
-check_btf:
-	/* check whether we recorded this BTF (and maybe module) already */
-	for (i = 0; i < env->used_btf_cnt; i++) {
-		if (env->used_btfs[i].btf == btf) {
-			btf_put(btf);
-			return 0;
-		}
-	}
-
-	if (env->used_btf_cnt >= MAX_USED_BTFS) {
-		err = -E2BIG;
-		goto err_put;
-	}
-
-	btf_mod = &env->used_btfs[env->used_btf_cnt];
-	btf_mod->btf = btf;
-	btf_mod->module = NULL;
-
-	/* if we reference variables from kernel module, bump its refcount */
-	if (btf_is_module(btf)) {
-		btf_mod->module = btf_try_get_module(btf);
-		if (!btf_mod->module) {
-			err = -ENXIO;
-			goto err_put;
-		}
-	}
-
-	env->used_btf_cnt++;
 
 	return 0;
-err_put:
-	btf_put(btf);
-	return err;
+}
+
+static int check_pseudo_btf_id(struct bpf_verifier_env *env,
+			       struct bpf_insn *insn,
+			       struct bpf_insn_aux_data *aux)
+{
+	struct btf *btf;
+	int btf_fd;
+	int err;
+
+	btf_fd = insn[1].imm;
+	if (btf_fd) {
+		CLASS(fd, f)(btf_fd);
+
+		btf = __btf_get_by_fd(f);
+		if (IS_ERR(btf)) {
+			verbose(env, "invalid module BTF object FD specified.\n");
+			return -EINVAL;
+		}
+	} else {
+		if (!btf_vmlinux) {
+			verbose(env, "kernel is missing BTF, make sure CONFIG_DEBUG_INFO_BTF=y is specified in Kconfig.\n");
+			return -EINVAL;
+		}
+		btf = btf_vmlinux;
+	}
+
+	err = __check_pseudo_btf_id(env, insn, aux, btf);
+	if (err)
+		return err;
+
+	err = __add_used_btf(env, btf);
+	if (err < 0)
+		return err;
+	return 0;
 }
 
 static bool is_tracing_prog_type(enum bpf_prog_type type)
···
 	default:
 		return false;
 	}
+}
+
+static bool bpf_map_is_cgroup_storage(struct bpf_map *map)
+{
+	return (map->map_type == BPF_MAP_TYPE_CGROUP_STORAGE ||
+		map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE);
 }
 
 static int check_map_prog_compatibility(struct bpf_verifier_env *env,
···
 			return -EINVAL;
 		}
 
+	if (bpf_map_is_cgroup_storage(map) &&
+	    bpf_cgroup_storage_assign(env->prog->aux, map)) {
+		verbose(env, "only one cgroup storage of each type is allowed\n");
+		return -EBUSY;
+	}
+
+	if (map->map_type == BPF_MAP_TYPE_ARENA) {
+		if (env->prog->aux->arena) {
+			verbose(env, "Only one arena per program\n");
+			return -EBUSY;
+		}
+		if (!env->allow_ptr_leaks || !env->bpf_capable) {
+			verbose(env, "CAP_BPF and CAP_PERFMON are required to use arena\n");
+			return -EPERM;
+		}
+		if (!env->prog->jit_requested) {
+			verbose(env, "JIT is required to use arena\n");
+			return -EOPNOTSUPP;
+		}
+		if (!bpf_jit_supports_arena()) {
+			verbose(env, "JIT doesn't support arena\n");
+			return -EOPNOTSUPP;
+		}
+		env->prog->aux->arena = (void *)map;
+		if (!bpf_arena_get_user_vm_start(env->prog->aux->arena)) {
+			verbose(env, "arena's user address must be set via map_extra or mmap()\n");
+			return -EINVAL;
+		}
+	}
+
 	return 0;
 }
 
-static bool bpf_map_is_cgroup_storage(struct bpf_map *map)
+static int __add_used_map(struct bpf_verifier_env *env, struct bpf_map *map)
 {
-	return (map->map_type == BPF_MAP_TYPE_CGROUP_STORAGE ||
-		map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE);
-}
-
-/* Add map behind fd to used maps list, if it's not already there, and return
- * its index. Also set *reused to true if this map was already in the list of
- * used maps.
- * Returns <0 on error, or >= 0 index, on success.
- */
-static int add_used_map_from_fd(struct bpf_verifier_env *env, int fd, bool *reused)
-{
-	CLASS(fd, f)(fd);
-	struct bpf_map *map;
-	int i;
-
-	map = __bpf_map_get(f);
-	if (IS_ERR(map)) {
-		verbose(env, "fd %d is not pointing to valid bpf_map\n", fd);
-		return PTR_ERR(map);
-	}
+	int i, err;
 
 	/* check whether we recorded this map already */
-	for (i = 0; i < env->used_map_cnt; i++) {
-		if (env->used_maps[i] == map) {
-			*reused = true;
+	for (i = 0; i < env->used_map_cnt; i++)
+		if (env->used_maps[i] == map)
 			return i;
-		}
-	}
 
 	if (env->used_map_cnt >= MAX_USED_MAPS) {
 		verbose(env, "The total number of maps per program has reached the limit of %u\n",
 			MAX_USED_MAPS);
 		return -E2BIG;
 	}
+
+	err = check_map_prog_compatibility(env, map, env->prog);
+	if (err)
+		return err;
 
 	if (env->prog->sleepable)
 		atomic64_inc(&map->sleepable_refcnt);
···
 	 */
 	bpf_map_inc(map);
 
-	*reused = false;
 	env->used_maps[env->used_map_cnt++] = map;
 
 	return env->used_map_cnt - 1;
+}
+
+/* Add map behind fd to used maps list, if it's not already there, and return
+ * its index.
+ * Returns <0 on error, or >= 0 index, on success.
+ */
+static int add_used_map(struct bpf_verifier_env *env, int fd)
+{
+	struct bpf_map *map;
+	CLASS(fd, f)(fd);
+
+	map = __bpf_map_get(f);
+	if (IS_ERR(map)) {
+		verbose(env, "fd %d is not pointing to valid bpf_map\n", fd);
+		return PTR_ERR(map);
+	}
+
+	return __add_used_map(env, map);
 }
 
 /* find and rewrite pseudo imm in ld_imm64 instructions:
···
 			int map_idx;
 			u64 addr;
 			u32 fd;
-			bool reused;
 
 			if (i == insn_cnt - 1 || insn[1].code != 0 ||
 			    insn[1].dst_reg != 0 || insn[1].src_reg != 0 ||
···
 				break;
 			}
 
-			map_idx = add_used_map_from_fd(env, fd, &reused);
+			map_idx = add_used_map(env, fd);
 			if (map_idx < 0)
 				return map_idx;
 			map = env->used_maps[map_idx];
 
 			aux = &env->insn_aux_data[i];
 			aux->map_index = map_idx;
-
-			err = check_map_prog_compatibility(env, map, env->prog);
-			if (err)
-				return err;
 
 			if (insn[0].src_reg == BPF_PSEUDO_MAP_FD ||
 			    insn[0].src_reg == BPF_PSEUDO_MAP_IDX) {
···
 
 			insn[0].imm = (u32)addr;
 			insn[1].imm = addr >> 32;
-
-			/* proceed with extra checks only if its newly added used map */
-			if (reused)
-				goto next_insn;
-
-			if (bpf_map_is_cgroup_storage(map) &&
-			    bpf_cgroup_storage_assign(env->prog->aux, map)) {
-				verbose(env, "only one cgroup storage of each type is allowed\n");
-				return -EBUSY;
-			}
-			if (map->map_type == BPF_MAP_TYPE_ARENA) {
-				if (env->prog->aux->arena) {
-					verbose(env, "Only one arena per program\n");
-					return -EBUSY;
-				}
-				if (!env->allow_ptr_leaks || !env->bpf_capable) {
-					verbose(env, "CAP_BPF and CAP_PERFMON are required to use arena\n");
-					return -EPERM;
-				}
-				if (!env->prog->jit_requested) {
-					verbose(env, "JIT is required to use arena\n");
-					return -EOPNOTSUPP;
-				}
-				if (!bpf_jit_supports_arena()) {
-					verbose(env, "JIT doesn't support arena\n");
-					return -EOPNOTSUPP;
-				}
-				env->prog->aux->arena = (void *)map;
-				if (!bpf_arena_get_user_vm_start(env->prog->aux->arena)) {
-					verbose(env, "arena's user address must be set via map_extra or mmap()\n");
-					return -EINVAL;
-				}
-			}
 
 next_insn:
 			insn++;
···
 	return btf_vmlinux;
 }
 
+/*
+ * The add_fd_from_fd_array() is executed only if fd_array_cnt is non-zero. In
+ * this case expect that every file descriptor in the array is either a map or
+ * a BTF. Everything else is considered to be trash.
+ */
+static int add_fd_from_fd_array(struct bpf_verifier_env *env, int fd)
+{
+	struct bpf_map *map;
+	struct btf *btf;
+	CLASS(fd, f)(fd);
+	int err;
+
+	map = __bpf_map_get(f);
+	if (!IS_ERR(map)) {
+		err = __add_used_map(env, map);
+		if (err < 0)
+			return err;
+		return 0;
+	}
+
+	btf = __btf_get_by_fd(f);
+	if (!IS_ERR(btf)) {
+		err = __add_used_btf(env, btf);
+		if (err < 0)
+			return err;
+		return 0;
+	}
+
+	verbose(env, "fd %d is not pointing to valid bpf_map or btf\n", fd);
+	return PTR_ERR(map);
+}
+
+static int process_fd_array(struct bpf_verifier_env *env, union bpf_attr *attr, bpfptr_t uattr)
+{
+	size_t size = sizeof(int);
+	int ret;
+	int fd;
+	u32 i;
+
+	env->fd_array = make_bpfptr(attr->fd_array, uattr.is_kernel);
+
+	/*
+	 * The only difference between old (no fd_array_cnt is given) and new
+	 * APIs is that in the latter case the fd_array is expected to be
+	 * continuous and is scanned for map fds right away
+	 */
+	if (!attr->fd_array_cnt)
+		return 0;
+
+	/* Check for integer overflow */
+	if (attr->fd_array_cnt >= (U32_MAX / size)) {
+		verbose(env, "fd_array_cnt is too big (%u)\n", attr->fd_array_cnt);
+		return -EINVAL;
+	}
+
+	for (i = 0; i < attr->fd_array_cnt; i++) {
+		if (copy_from_bpfptr_offset(&fd, env->fd_array, i * size, size))
+			return -EFAULT;
+
+		ret = add_fd_from_fd_array(env, fd);
+		if (ret)
+			return ret;
+	}
+
+	return 0;
+}
+
 int bpf_check(struct bpf_prog **prog, union bpf_attr *attr, bpfptr_t uattr, __u32 uattr_size)
 {
 	u64 start_time = ktime_get_ns();
···
 		env->insn_aux_data[i].orig_idx = i;
 	env->prog = *prog;
 	env->ops = bpf_verifier_ops[env->prog->type];
-	env->fd_array = make_bpfptr(attr->fd_array, uattr.is_kernel);
 
 	env->allow_ptr_leaks = bpf_allow_ptr_leaks(env->prog->aux->token);
 	env->allow_uninit_stack = bpf_allow_uninit_stack(env->prog->aux->token);
···
 			    attr->log_size);
 	if (ret)
 		goto err_unlock;
+
+	ret = process_fd_array(env, attr, uattr);
+	if (ret)
+		goto skip_full_check;
 
 	mark_verifier_state_clean(env);
 

tools/include/uapi/linux/bpf.h

···
 		 * If provided, prog_flags should have BPF_F_TOKEN_FD flag set.
 		 */
 		__s32		prog_token_fd;
+		/* The fd_array_cnt can be used to pass the length of the
+		 * fd_array array. In this case all the [map] file descriptors
+		 * passed in this array will be bound to the program, even if
+		 * the maps are not referenced directly. The functionality is
+		 * similar to the BPF_PROG_BIND_MAP syscall, but maps can be
+		 * used by the verifier during the program load. If provided,
+		 * then the fd_array[0,...,fd_array_cnt-1] is expected to be
+		 * continuous.
+		 */
+		__u32		fd_array_cnt;
 	};
 
 	struct { /* anonymous struct used by BPF_OBJ_* commands */

tools/lib/bpf/bpf.c

···
 		  const struct bpf_insn *insns, size_t insn_cnt,
 		  struct bpf_prog_load_opts *opts)
 {
-	const size_t attr_sz = offsetofend(union bpf_attr, prog_token_fd);
+	const size_t attr_sz = offsetofend(union bpf_attr, fd_array_cnt);
 	void *finfo = NULL, *linfo = NULL;
 	const char *func_info, *line_info;
 	__u32 log_size, log_level, attach_prog_fd, attach_btf_obj_fd;
···
 	attr.line_info_cnt = OPTS_GET(opts, line_info_cnt, 0);
 
 	attr.fd_array = ptr_to_u64(OPTS_GET(opts, fd_array, NULL));
+	attr.fd_array_cnt = OPTS_GET(opts, fd_array_cnt, 0);
 
 	if (log_level) {
 		attr.log_buf = ptr_to_u64(log_buf);

tools/lib/bpf/bpf.h

···
 	 */
 	__u32 log_true_size;
 	__u32 token_fd;
+
+	/* if set, provides the length of fd_array */
+	__u32 fd_array_cnt;
 	size_t :0;
 };
-#define bpf_prog_load_opts__last_field token_fd
+#define bpf_prog_load_opts__last_field fd_array_cnt
 
 LIBBPF_API int bpf_prog_load(enum bpf_prog_type prog_type,
 			     const char *prog_name, const char *license,

tools/testing/selftests/bpf/prog_tests/fd_array.c

···
+// SPDX-License-Identifier: GPL-2.0
+
+#include <test_progs.h>
+
+#include <linux/btf.h>
+#include <bpf/bpf.h>
+
+#include "../test_btf.h"
+
+static inline int new_map(void)
+{
+	const char *name = NULL;
+	__u32 max_entries = 1;
+	__u32 value_size = 8;
+	__u32 key_size = 4;
+
+	return bpf_map_create(BPF_MAP_TYPE_ARRAY, name,
+			      key_size, value_size,
+			      max_entries, NULL);
+}
+
+static int new_btf(void)
+{
+	struct btf_blob {
+		struct btf_header btf_hdr;
+		__u32 types[8];
+		__u32 str;
+	} raw_btf = {
+		.btf_hdr = {
+			.magic = BTF_MAGIC,
+			.version = BTF_VERSION,
+			.hdr_len = sizeof(struct btf_header),
+			.type_len = sizeof(raw_btf.types),
+			.str_off = offsetof(struct btf_blob, str) - offsetof(struct btf_blob, types),
+			.str_len = sizeof(raw_btf.str),
+		},
+		.types = {
+			/* long */
+			BTF_TYPE_INT_ENC(0, BTF_INT_SIGNED, 0, 64, 8),  /* [1] */
+			/* unsigned long */
+			BTF_TYPE_INT_ENC(0, 0, 0, 64, 8),  /* [2] */
+		},
+	};
+
+	return bpf_btf_load(&raw_btf, sizeof(raw_btf), NULL);
+}
+
+#define Close(FD) do {		\
+	if ((FD) >= 0) {	\
+		close(FD);	\
+		FD = -1;	\
+	}			\
+} while(0)
+
+static bool map_exists(__u32 id)
+{
+	int fd;
+
+	fd = bpf_map_get_fd_by_id(id);
+	if (fd >= 0) {
+		close(fd);
+		return true;
+	}
+	return false;
+}
+
+static bool btf_exists(__u32 id)
+{
+	int fd;
+
+	fd = bpf_btf_get_fd_by_id(id);
+	if (fd >= 0) {
+		close(fd);
+		return true;
+	}
+	return false;
+}
+
+static inline int bpf_prog_get_map_ids(int prog_fd, __u32 *nr_map_ids, __u32 *map_ids)
+{
+	__u32 len = sizeof(struct bpf_prog_info);
+	struct bpf_prog_info info;
+	int err;
+
+	memset(&info, 0, len);
+	info.nr_map_ids = *nr_map_ids,
+	info.map_ids = ptr_to_u64(map_ids),
+
+	err = bpf_prog_get_info_by_fd(prog_fd, &info, &len);
+	if (!ASSERT_OK(err, "bpf_prog_get_info_by_fd"))
+		return -1;
+
+	*nr_map_ids = info.nr_map_ids;
+
+	return 0;
+}
+
+static int __load_test_prog(int map_fd, const int *fd_array, int fd_array_cnt)
+{
+	/* A trivial program which uses one map */
+	struct bpf_insn insns[] = {
+		BPF_LD_MAP_FD(BPF_REG_1, map_fd),
+		BPF_ST_MEM(BPF_DW, BPF_REG_10, -8, 0),
+		BPF_MOV64_REG(BPF_REG_2, BPF_REG_10),
+		BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -8),
+		BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem),
+		BPF_MOV64_IMM(BPF_REG_0, 0),
+		BPF_EXIT_INSN(),
+	};
+	LIBBPF_OPTS(bpf_prog_load_opts, opts);
+
+	opts.fd_array = fd_array;
+	opts.fd_array_cnt = fd_array_cnt;
+
+	return bpf_prog_load(BPF_PROG_TYPE_XDP, NULL, "GPL", insns, ARRAY_SIZE(insns), &opts);
+}
+
+static int load_test_prog(const int *fd_array, int fd_array_cnt)
+{
+	int map_fd;
+	int ret;
+
+	map_fd = new_map();
+	if (!ASSERT_GE(map_fd, 0, "new_map"))
+		return map_fd;
+
+	ret = __load_test_prog(map_fd, fd_array, fd_array_cnt);
+	close(map_fd);
+	return ret;
+}
+
+static bool check_expected_map_ids(int prog_fd, int expected, __u32 *map_ids, __u32 *nr_map_ids)
+{
+	int err;
+
+	err = bpf_prog_get_map_ids(prog_fd, nr_map_ids, map_ids);
+	if (!ASSERT_OK(err, "bpf_prog_get_map_ids"))
+		return false;
+	if (!ASSERT_EQ(*nr_map_ids, expected, "unexpected nr_map_ids"))
+		return false;
+
+	return true;
+}
+
+/*
+ * Load a program, which uses one map. No fd_array maps are present.
+ * On return only one map is expected to be bound to prog.
+ */
+static void check_fd_array_cnt__no_fd_array(void)
+{
+	__u32 map_ids[16];
+	__u32 nr_map_ids;
+	int prog_fd = -1;
+
+	prog_fd = load_test_prog(NULL, 0);
+	if (!ASSERT_GE(prog_fd, 0, "BPF_PROG_LOAD"))
+		return;
+	nr_map_ids = ARRAY_SIZE(map_ids);
+	check_expected_map_ids(prog_fd, 1, map_ids, &nr_map_ids);
+	close(prog_fd);
+}
+
+/*
+ * Load a program, which uses one map, and pass two extra, non-equal, maps in
+ * fd_array with fd_array_cnt=2. On return three maps are expected to be bound
+ * to the program.
+ */
+static void check_fd_array_cnt__fd_array_ok(void)
+{
+	int extra_fds[2] = { -1, -1 };
+	__u32 map_ids[16];
+	__u32 nr_map_ids;
+	int prog_fd = -1;
+
+	extra_fds[0] = new_map();
+	if (!ASSERT_GE(extra_fds[0], 0, "new_map"))
+		goto cleanup;
+	extra_fds[1] = new_map();
+	if (!ASSERT_GE(extra_fds[1], 0, "new_map"))
+		goto cleanup;
+	prog_fd = load_test_prog(extra_fds, 2);
+	if (!ASSERT_GE(prog_fd, 0, "BPF_PROG_LOAD"))
+		goto cleanup;
+	nr_map_ids = ARRAY_SIZE(map_ids);
+	if (!check_expected_map_ids(prog_fd, 3, map_ids, &nr_map_ids))
+		goto cleanup;
+
+	/* maps should still exist when original file descriptors are closed */
+	Close(extra_fds[0]);
+	Close(extra_fds[1]);
+	if (!ASSERT_EQ(map_exists(map_ids[0]), true, "map_ids[0] should exist"))
+		goto cleanup;
+	if (!ASSERT_EQ(map_exists(map_ids[1]), true, "map_ids[1] should exist"))
+		goto cleanup;
+
+	/* some fds might be invalid, so ignore return codes */
+cleanup:
+	Close(extra_fds[1]);
+	Close(extra_fds[0]);
+	Close(prog_fd);
+}
+
+/*
+ * Load a program with a few extra maps duplicated in the fd_array.
+ * After the load maps should only be referenced once.
+ */
+static void check_fd_array_cnt__duplicated_maps(void)
+{
+	int extra_fds[4] = { -1, -1, -1, -1 };
+	__u32 map_ids[16];
+	__u32 nr_map_ids;
+	int prog_fd = -1;
+
+	extra_fds[0] = extra_fds[2] = new_map();
+	if (!ASSERT_GE(extra_fds[0], 0, "new_map"))
+		goto cleanup;
+	extra_fds[1] = extra_fds[3] = new_map();
+	if (!ASSERT_GE(extra_fds[1], 0, "new_map"))
+		goto cleanup;
+	prog_fd = load_test_prog(extra_fds, 4);
+	if (!ASSERT_GE(prog_fd, 0, "BPF_PROG_LOAD"))
+		goto cleanup;
+	nr_map_ids = ARRAY_SIZE(map_ids);
+	if (!check_expected_map_ids(prog_fd, 3, map_ids, &nr_map_ids))
+		goto cleanup;
+
+	/* maps should still exist when original file descriptors are closed */
+	Close(extra_fds[0]);
+	Close(extra_fds[1]);
+	if (!ASSERT_EQ(map_exists(map_ids[0]), true, "map should exist"))
+		goto cleanup;
+	if (!ASSERT_EQ(map_exists(map_ids[1]), true, "map should exist"))
+		goto cleanup;
+
+	/* some fds might be invalid, so ignore return codes */
+cleanup:
+	Close(extra_fds[1]);
+	Close(extra_fds[0]);
+	Close(prog_fd);
+}
+
+/*
+ * Check that if maps which are referenced by a program are
+ * passed in fd_array, then they will be referenced only once
+ */
+static void check_fd_array_cnt__referenced_maps_in_fd_array(void)
+{
+	int extra_fds[1] = { -1 };
+	__u32 map_ids[16];
+	__u32 nr_map_ids;
+	int prog_fd = -1;
+
+	extra_fds[0] = new_map();
+	if (!ASSERT_GE(extra_fds[0], 0, "new_map"))
+		goto cleanup;
+	prog_fd = __load_test_prog(extra_fds[0], extra_fds, 1);
+	if (!ASSERT_GE(prog_fd, 0, "BPF_PROG_LOAD"))
+		goto cleanup;
+	nr_map_ids = ARRAY_SIZE(map_ids);
+	if (!check_expected_map_ids(prog_fd, 1, map_ids, &nr_map_ids))
+		goto cleanup;
+
+	/* map should still exist when original file descriptor is closed */
+	Close(extra_fds[0]);
+	if (!ASSERT_EQ(map_exists(map_ids[0]), true, "map should exist"))
+		goto cleanup;
+
+	/* some fds might be invalid, so ignore return codes */
+cleanup:
+	Close(extra_fds[0]);
+	Close(prog_fd);
+}
+
+static int get_btf_id_by_fd(int btf_fd, __u32 *id)
+{
+	struct bpf_btf_info info;
+	__u32 info_len = sizeof(info);
+	int err;
+
+	memset(&info, 0, info_len);
+	err = bpf_btf_get_info_by_fd(btf_fd, &info, &info_len);
+	if (err)
+		return err;
+	if (id)
+		*id = info.id;
+	return 0;
+}
+
+/*
+ * Check that fd_array operates properly for btfs. Namely, to check that
+ * passing a btf fd in fd_array increases its reference count, do the
+ * following:
+ *  1) Create a new btf, it's referenced only by a file descriptor, so refcnt=1
+ *  2) Load a BPF prog with fd_array[0] = btf_fd; now btf's refcnt=2
+ *  3) Close the btf_fd, now refcnt=1
+ * Wait and check that BTF stil exists.
+ */
+static void check_fd_array_cnt__referenced_btfs(void)
+{
+	int extra_fds[1] = { -1 };
+	int prog_fd = -1;
+	__u32 btf_id;
+	int tries;
+	int err;
+
+	extra_fds[0] = new_btf();
+	if (!ASSERT_GE(extra_fds[0], 0, "new_btf"))
+		goto cleanup;
+	prog_fd = load_test_prog(extra_fds, 1);
+	if (!ASSERT_GE(prog_fd, 0, "BPF_PROG_LOAD"))
+		goto cleanup;
+
+	/* btf should still exist when original file descriptor is closed */
+	err = get_btf_id_by_fd(extra_fds[0], &btf_id);
+	if (!ASSERT_GE(err, 0, "get_btf_id_by_fd"))
+		goto cleanup;
+
+	Close(extra_fds[0]);
+
+	if (!ASSERT_GE(kern_sync_rcu(), 0, "kern_sync_rcu 1"))
+		goto cleanup;
+
+	if (!ASSERT_EQ(btf_exists(btf_id), true, "btf should exist"))
+		goto cleanup;
+
+	Close(prog_fd);
+
+	/* The program is freed by a workqueue, so no reliable
+	 * way to sync, so just wait a bit (max ~1 second). */
+	for (tries = 100; tries >= 0; tries--) {
+		usleep(1000);
+
+		if (!btf_exists(btf_id))
+			break;
+
+		if (tries)
+			continue;
+
+		PRINT_FAIL("btf should have been freed");
+	}
+
+	/* some fds might be invalid, so ignore return codes */
+cleanup:
+	Close(extra_fds[0]);
+	Close(prog_fd);
+}
+
+/*
+ * Test that a program with trash in fd_array can't be loaded:
+ * only map and BTF file descriptors should be accepted.
+ */
+static void check_fd_array_cnt__fd_array_with_trash(void)
+{
+	int extra_fds[3] = { -1, -1, -1 };
+	int prog_fd = -1;
+
+	extra_fds[0] = new_map();
+	if (!ASSERT_GE(extra_fds[0], 0, "new_map"))
+		goto cleanup;
+	extra_fds[1] = new_btf();
+	if (!ASSERT_GE(extra_fds[1], 0, "new_btf"))
+		goto cleanup;
+
+	/* trash 1: not a file descriptor */
+	extra_fds[2] = 0xbeef;
+	prog_fd = load_test_prog(extra_fds, 3);
+	if (!ASSERT_EQ(prog_fd, -EBADF, "prog should have been rejected with -EBADF"))
+		goto cleanup;
+
+	/* trash 2: not a map or btf */
+	extra_fds[2] = socket(AF_INET, SOCK_STREAM, 0);
+	if (!ASSERT_GE(extra_fds[2], 0, "socket"))
+		goto cleanup;
+
+	prog_fd = load_test_prog(extra_fds, 3);
+	if (!ASSERT_EQ(prog_fd, -EINVAL, "prog should have been rejected with -EINVAL"))
+		goto cleanup;
+
+	/* Validate that the prog is ok if trash is removed */
+	Close(extra_fds[2]);
+	extra_fds[2] = new_btf();
+	if (!ASSERT_GE(extra_fds[2], 0, "new_btf"))
+		goto cleanup;
+
+	prog_fd = load_test_prog(extra_fds, 3);
+	if (!ASSERT_GE(prog_fd, 0, "prog should have been loaded"))
+		goto cleanup;
+
+	/* some fds might be invalid, so ignore return codes */
+cleanup:
+	Close(extra_fds[2]);
+	Close(extra_fds[1]);
+	Close(extra_fds[0]);
+}
+
+/*
+ * Test that a program with too big fd_array can't be loaded.
+ */
+static void check_fd_array_cnt__fd_array_too_big(void)
+{
+	int extra_fds[65];
+	int prog_fd = -1;
+	int i;
+
+	for (i = 0; i < 65; i++) {
+		extra_fds[i] = new_map();
+		if (!ASSERT_GE(extra_fds[i], 0, "new_map"))
+			goto cleanup_fds;
+	}
+
+	prog_fd = load_test_prog(extra_fds, 65);
+	ASSERT_EQ(prog_fd, -E2BIG, "prog should have been rejected with -E2BIG");
+
+cleanup_fds:
+	while (i > 0)
+		Close(extra_fds[--i]);
+}
+
+void test_fd_array_cnt(void)
+{
+	if (test__start_subtest("no-fd-array"))
+		check_fd_array_cnt__no_fd_array();
+
+	if (test__start_subtest("fd-array-ok"))
+		check_fd_array_cnt__fd_array_ok();
+
+	if (test__start_subtest("fd-array-dup-input"))
+		check_fd_array_cnt__duplicated_maps();
+
+	if (test__start_subtest("fd-array-ref-maps-in-array"))
+		check_fd_array_cnt__referenced_maps_in_fd_array();
+
+	if (test__start_subtest("fd-array-ref-btfs"))
+		check_fd_array_cnt__referenced_btfs();
+
+	if (test__start_subtest("fd-array-trash-input"))
+		check_fd_array_cnt__fd_array_with_trash();
+
+	if (test__start_subtest("fd-array-2big"))
+		check_fd_array_cnt__fd_array_too_big();
+}

tools/testing/selftests/bpf/progs/syscall.c

···
 			.magic = BTF_MAGIC,
 			.version = BTF_VERSION,
 			.hdr_len = sizeof(struct btf_header),
-			.type_len = sizeof(__u32) * 8,
-			.str_off = sizeof(__u32) * 8,
-			.str_len = sizeof(__u32),
+			.type_len = sizeof(raw_btf.types),
+			.str_off = offsetof(struct btf_blob, str) - offsetof(struct btf_blob, types),
+			.str_len = sizeof(raw_btf.str),
 		},
 		.types = {
 			/* long */