1 // SPDX-License-Identifier: GPL-2.0-only
3 * Minimal file system backend for holding eBPF maps and programs,
4 * used by bpf(2) object pinning.
8 * Daniel Borkmann <daniel@iogearbox.net>
11 #include <linux/init.h>
12 #include <linux/magic.h>
13 #include <linux/major.h>
14 #include <linux/mount.h>
15 #include <linux/namei.h>
17 #include <linux/fs_context.h>
18 #include <linux/fs_parser.h>
19 #include <linux/kdev_t.h>
20 #include <linux/filter.h>
21 #include <linux/bpf.h>
22 #include <linux/bpf_trace.h>
30 static void *bpf_any_get(void *raw, enum bpf_type type)
34 raw = bpf_prog_inc(raw);
37 raw = bpf_map_inc(raw, true);
47 static void bpf_any_put(void *raw, enum bpf_type type)
54 bpf_map_put_with_uref(raw);
62 static void *bpf_fd_probe_obj(u32 ufd, enum bpf_type *type)
67 raw = bpf_map_get_with_uref(ufd);
69 *type = BPF_TYPE_PROG;
70 raw = bpf_prog_get(ufd);
76 static const struct inode_operations bpf_dir_iops;
78 static const struct inode_operations bpf_prog_iops = { };
79 static const struct inode_operations bpf_map_iops = { };
81 static struct inode *bpf_get_inode(struct super_block *sb,
82 const struct inode *dir,
87 switch (mode & S_IFMT) {
93 return ERR_PTR(-EINVAL);
96 inode = new_inode(sb);
98 return ERR_PTR(-ENOSPC);
100 inode->i_ino = get_next_ino();
101 inode->i_atime = current_time(inode);
102 inode->i_mtime = inode->i_atime;
103 inode->i_ctime = inode->i_atime;
105 inode_init_owner(inode, dir, mode);
110 static int bpf_inode_type(const struct inode *inode, enum bpf_type *type)
112 *type = BPF_TYPE_UNSPEC;
113 if (inode->i_op == &bpf_prog_iops)
114 *type = BPF_TYPE_PROG;
115 else if (inode->i_op == &bpf_map_iops)
116 *type = BPF_TYPE_MAP;
123 static void bpf_dentry_finalize(struct dentry *dentry, struct inode *inode,
126 d_instantiate(dentry, inode);
129 dir->i_mtime = current_time(dir);
130 dir->i_ctime = dir->i_mtime;
133 static int bpf_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
137 inode = bpf_get_inode(dir->i_sb, dir, mode | S_IFDIR);
139 return PTR_ERR(inode);
141 inode->i_op = &bpf_dir_iops;
142 inode->i_fop = &simple_dir_operations;
147 bpf_dentry_finalize(dentry, inode, dir);
156 static struct map_iter *map_iter(struct seq_file *m)
161 static struct bpf_map *seq_file_to_map(struct seq_file *m)
163 return file_inode(m->file)->i_private;
166 static void map_iter_free(struct map_iter *iter)
174 static struct map_iter *map_iter_alloc(struct bpf_map *map)
176 struct map_iter *iter;
178 iter = kzalloc(sizeof(*iter), GFP_KERNEL | __GFP_NOWARN);
182 iter->key = kzalloc(map->key_size, GFP_KERNEL | __GFP_NOWARN);
193 static void *map_seq_next(struct seq_file *m, void *v, loff_t *pos)
195 struct bpf_map *map = seq_file_to_map(m);
196 void *key = map_iter(m)->key;
200 if (map_iter(m)->done)
203 if (unlikely(v == SEQ_START_TOKEN))
209 if (map->ops->map_get_next_key(map, prev_key, key)) {
210 map_iter(m)->done = true;
217 static void *map_seq_start(struct seq_file *m, loff_t *pos)
219 if (map_iter(m)->done)
222 return *pos ? map_iter(m)->key : SEQ_START_TOKEN;
225 static void map_seq_stop(struct seq_file *m, void *v)
229 static int map_seq_show(struct seq_file *m, void *v)
231 struct bpf_map *map = seq_file_to_map(m);
232 void *key = map_iter(m)->key;
234 if (unlikely(v == SEQ_START_TOKEN)) {
235 seq_puts(m, "# WARNING!! The output is for debug purpose only\n");
236 seq_puts(m, "# WARNING!! The output format will change\n");
238 map->ops->map_seq_show_elem(map, key, m);
244 static const struct seq_operations bpffs_map_seq_ops = {
245 .start = map_seq_start,
246 .next = map_seq_next,
247 .show = map_seq_show,
248 .stop = map_seq_stop,
251 static int bpffs_map_open(struct inode *inode, struct file *file)
253 struct bpf_map *map = inode->i_private;
254 struct map_iter *iter;
258 iter = map_iter_alloc(map);
262 err = seq_open(file, &bpffs_map_seq_ops);
268 m = file->private_data;
274 static int bpffs_map_release(struct inode *inode, struct file *file)
276 struct seq_file *m = file->private_data;
278 map_iter_free(map_iter(m));
280 return seq_release(inode, file);
283 /* bpffs_map_fops should only implement the basic
284 * read operation for a BPF map. The purpose is to
285 * provide a simple user intuitive way to do
286 * "cat bpffs/pathto/a-pinned-map".
288 * Other operations (e.g. write, lookup...) should be realized by
289 * the userspace tools (e.g. bpftool) through the
290 * BPF_OBJ_GET_INFO_BY_FD and the map's lookup/update
293 static const struct file_operations bpffs_map_fops = {
294 .open = bpffs_map_open,
296 .release = bpffs_map_release,
299 static int bpffs_obj_open(struct inode *inode, struct file *file)
304 static const struct file_operations bpffs_obj_fops = {
305 .open = bpffs_obj_open,
308 static int bpf_mkobj_ops(struct dentry *dentry, umode_t mode, void *raw,
309 const struct inode_operations *iops,
310 const struct file_operations *fops)
312 struct inode *dir = dentry->d_parent->d_inode;
313 struct inode *inode = bpf_get_inode(dir->i_sb, dir, mode);
315 return PTR_ERR(inode);
319 inode->i_private = raw;
321 bpf_dentry_finalize(dentry, inode, dir);
325 static int bpf_mkprog(struct dentry *dentry, umode_t mode, void *arg)
327 return bpf_mkobj_ops(dentry, mode, arg, &bpf_prog_iops,
331 static int bpf_mkmap(struct dentry *dentry, umode_t mode, void *arg)
333 struct bpf_map *map = arg;
335 return bpf_mkobj_ops(dentry, mode, arg, &bpf_map_iops,
336 bpf_map_support_seq_show(map) ?
337 &bpffs_map_fops : &bpffs_obj_fops);
340 static struct dentry *
341 bpf_lookup(struct inode *dir, struct dentry *dentry, unsigned flags)
343 /* Dots in names (e.g. "/sys/fs/bpf/foo.bar") are reserved for future
346 if (strchr(dentry->d_name.name, '.'))
347 return ERR_PTR(-EPERM);
349 return simple_lookup(dir, dentry, flags);
352 static int bpf_symlink(struct inode *dir, struct dentry *dentry,
355 char *link = kstrdup(target, GFP_USER | __GFP_NOWARN);
361 inode = bpf_get_inode(dir->i_sb, dir, S_IRWXUGO | S_IFLNK);
364 return PTR_ERR(inode);
367 inode->i_op = &simple_symlink_inode_operations;
368 inode->i_link = link;
370 bpf_dentry_finalize(dentry, inode, dir);
374 static const struct inode_operations bpf_dir_iops = {
375 .lookup = bpf_lookup,
377 .symlink = bpf_symlink,
378 .rmdir = simple_rmdir,
379 .rename = simple_rename,
381 .unlink = simple_unlink,
384 static int bpf_obj_do_pin(const struct filename *pathname, void *raw,
387 struct dentry *dentry;
393 dentry = kern_path_create(AT_FDCWD, pathname->name, &path, 0);
395 return PTR_ERR(dentry);
397 mode = S_IFREG | ((S_IRUSR | S_IWUSR) & ~current_umask());
399 ret = security_path_mknod(&path, dentry, mode, 0);
403 dir = d_inode(path.dentry);
404 if (dir->i_op != &bpf_dir_iops) {
411 ret = vfs_mkobj(dentry, mode, bpf_mkprog, raw);
414 ret = vfs_mkobj(dentry, mode, bpf_mkmap, raw);
420 done_path_create(&path, dentry);
424 int bpf_obj_pin_user(u32 ufd, const char __user *pathname)
426 struct filename *pname;
431 pname = getname(pathname);
433 return PTR_ERR(pname);
435 raw = bpf_fd_probe_obj(ufd, &type);
441 ret = bpf_obj_do_pin(pname, raw, type);
443 bpf_any_put(raw, type);
449 static void *bpf_obj_do_get(const struct filename *pathname,
450 enum bpf_type *type, int flags)
457 ret = kern_path(pathname->name, LOOKUP_FOLLOW, &path);
461 inode = d_backing_inode(path.dentry);
462 ret = inode_permission(inode, ACC_MODE(flags));
466 ret = bpf_inode_type(inode, type);
470 raw = bpf_any_get(inode->i_private, *type);
481 int bpf_obj_get_user(const char __user *pathname, int flags)
483 enum bpf_type type = BPF_TYPE_UNSPEC;
484 struct filename *pname;
489 f_flags = bpf_get_file_flag(flags);
493 pname = getname(pathname);
495 return PTR_ERR(pname);
497 raw = bpf_obj_do_get(pname, &type, f_flags);
503 if (type == BPF_TYPE_PROG)
504 ret = bpf_prog_new_fd(raw);
505 else if (type == BPF_TYPE_MAP)
506 ret = bpf_map_new_fd(raw, f_flags);
511 bpf_any_put(raw, type);
517 static struct bpf_prog *__get_prog_inode(struct inode *inode, enum bpf_prog_type type)
519 struct bpf_prog *prog;
520 int ret = inode_permission(inode, MAY_READ);
524 if (inode->i_op == &bpf_map_iops)
525 return ERR_PTR(-EINVAL);
526 if (inode->i_op != &bpf_prog_iops)
527 return ERR_PTR(-EACCES);
529 prog = inode->i_private;
531 ret = security_bpf_prog(prog);
535 if (!bpf_prog_get_ok(prog, &type, false))
536 return ERR_PTR(-EINVAL);
538 return bpf_prog_inc(prog);
541 struct bpf_prog *bpf_prog_get_type_path(const char *name, enum bpf_prog_type type)
543 struct bpf_prog *prog;
545 int ret = kern_path(name, LOOKUP_FOLLOW, &path);
548 prog = __get_prog_inode(d_backing_inode(path.dentry), type);
554 EXPORT_SYMBOL(bpf_prog_get_type_path);
557 * Display the mount options in /proc/mounts.
559 static int bpf_show_options(struct seq_file *m, struct dentry *root)
561 umode_t mode = d_inode(root)->i_mode & S_IALLUGO & ~S_ISVTX;
563 if (mode != S_IRWXUGO)
564 seq_printf(m, ",mode=%o", mode);
568 static void bpf_free_inode(struct inode *inode)
572 if (S_ISLNK(inode->i_mode))
573 kfree(inode->i_link);
574 if (!bpf_inode_type(inode, &type))
575 bpf_any_put(inode->i_private, type);
576 free_inode_nonrcu(inode);
579 static const struct super_operations bpf_super_ops = {
580 .statfs = simple_statfs,
581 .drop_inode = generic_delete_inode,
582 .show_options = bpf_show_options,
583 .free_inode = bpf_free_inode,
590 static const struct fs_parameter_spec bpf_param_specs[] = {
591 fsparam_u32oct ("mode", OPT_MODE),
595 static const struct fs_parameter_description bpf_fs_parameters = {
597 .specs = bpf_param_specs,
600 struct bpf_mount_opts {
604 static int bpf_parse_param(struct fs_context *fc, struct fs_parameter *param)
606 struct bpf_mount_opts *opts = fc->fs_private;
607 struct fs_parse_result result;
610 opt = fs_parse(fc, &bpf_fs_parameters, param, &result);
612 /* We might like to report bad mount options here, but
613 * traditionally we've ignored all mount options, so we'd
614 * better continue to ignore non-existing options for bpf.
616 return opt == -ENOPARAM ? 0 : opt;
620 opts->mode = result.uint_32 & S_IALLUGO;
627 static int bpf_fill_super(struct super_block *sb, struct fs_context *fc)
629 static const struct tree_descr bpf_rfiles[] = { { "" } };
630 struct bpf_mount_opts *opts = fc->fs_private;
634 ret = simple_fill_super(sb, BPF_FS_MAGIC, bpf_rfiles);
638 sb->s_op = &bpf_super_ops;
640 inode = sb->s_root->d_inode;
641 inode->i_op = &bpf_dir_iops;
642 inode->i_mode &= ~S_IALLUGO;
643 inode->i_mode |= S_ISVTX | opts->mode;
648 static int bpf_get_tree(struct fs_context *fc)
650 return get_tree_nodev(fc, bpf_fill_super);
653 static void bpf_free_fc(struct fs_context *fc)
655 kfree(fc->fs_private);
658 static const struct fs_context_operations bpf_context_ops = {
660 .parse_param = bpf_parse_param,
661 .get_tree = bpf_get_tree,
665 * Set up the filesystem mount context.
667 static int bpf_init_fs_context(struct fs_context *fc)
669 struct bpf_mount_opts *opts;
671 opts = kzalloc(sizeof(struct bpf_mount_opts), GFP_KERNEL);
675 opts->mode = S_IRWXUGO;
677 fc->fs_private = opts;
678 fc->ops = &bpf_context_ops;
682 static struct file_system_type bpf_fs_type = {
683 .owner = THIS_MODULE,
685 .init_fs_context = bpf_init_fs_context,
686 .parameters = &bpf_fs_parameters,
687 .kill_sb = kill_litter_super,
690 static int __init bpf_init(void)
694 ret = sysfs_create_mount_point(fs_kobj, "bpf");
698 ret = register_filesystem(&bpf_fs_type);
700 sysfs_remove_mount_point(fs_kobj, "bpf");
704 fs_initcall(bpf_init);