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/[email protected]
Signed-off-by: Andrii Nakryiko <[email protected]>

Author: anakryiko

include/linux/bpf.h

@@ -2301,6 +2301,14 @@ void __bpf_obj_drop_impl(void *p, const struct btf_record *rec, bool percpu);
 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))
@@ -2310,6 +2318,15 @@ static inline struct bpf_map *__bpf_map_get(struct fd f)
 	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;
+}
+
 void bpf_map_inc(struct bpf_map *map);
 void bpf_map_inc_with_uref(struct bpf_map *map);
 struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map, bool uref);

include/uapi/linux/bpf.h

@@ -1573,6 +1573,16 @@ union bpf_attr {
 		 * 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

@@ -7746,14 +7746,9 @@ struct btf *btf_get_by_fd(int fd)
 	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

@@ -2730,7 +2730,7 @@ static bool is_perfmon_prog_type(enum bpf_prog_type prog_type)
 }
 
 /* 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)
 {