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>
23 #include "preload/bpf_preload.h"
32 static void *bpf_any_get(void *raw, enum bpf_type type)
39 bpf_map_inc_with_uref(raw);
52 static void bpf_any_put(void *raw, enum bpf_type type)
59 bpf_map_put_with_uref(raw);
70 static void *bpf_fd_probe_obj(u32 ufd, enum bpf_type *type)
74 raw = bpf_map_get_with_uref(ufd);
80 raw = bpf_prog_get(ufd);
82 *type = BPF_TYPE_PROG;
86 raw = bpf_link_get_from_fd(ufd);
88 *type = BPF_TYPE_LINK;
92 return ERR_PTR(-EINVAL);
95 static const struct inode_operations bpf_dir_iops;
97 static const struct inode_operations bpf_prog_iops = { };
98 static const struct inode_operations bpf_map_iops = { };
99 static const struct inode_operations bpf_link_iops = { };
101 static struct inode *bpf_get_inode(struct super_block *sb,
102 const struct inode *dir,
107 switch (mode & S_IFMT) {
113 return ERR_PTR(-EINVAL);
116 inode = new_inode(sb);
118 return ERR_PTR(-ENOSPC);
120 inode->i_ino = get_next_ino();
121 inode->i_atime = inode_set_ctime_current(inode);
122 inode->i_mtime = inode->i_atime;
124 inode_init_owner(&nop_mnt_idmap, inode, dir, mode);
129 static int bpf_inode_type(const struct inode *inode, enum bpf_type *type)
131 *type = BPF_TYPE_UNSPEC;
132 if (inode->i_op == &bpf_prog_iops)
133 *type = BPF_TYPE_PROG;
134 else if (inode->i_op == &bpf_map_iops)
135 *type = BPF_TYPE_MAP;
136 else if (inode->i_op == &bpf_link_iops)
137 *type = BPF_TYPE_LINK;
144 static void bpf_dentry_finalize(struct dentry *dentry, struct inode *inode,
147 d_instantiate(dentry, inode);
150 dir->i_mtime = inode_set_ctime_current(dir);
153 static int bpf_mkdir(struct mnt_idmap *idmap, struct inode *dir,
154 struct dentry *dentry, umode_t mode)
158 inode = bpf_get_inode(dir->i_sb, dir, mode | S_IFDIR);
160 return PTR_ERR(inode);
162 inode->i_op = &bpf_dir_iops;
163 inode->i_fop = &simple_dir_operations;
168 bpf_dentry_finalize(dentry, inode, dir);
177 static struct map_iter *map_iter(struct seq_file *m)
182 static struct bpf_map *seq_file_to_map(struct seq_file *m)
184 return file_inode(m->file)->i_private;
187 static void map_iter_free(struct map_iter *iter)
195 static struct map_iter *map_iter_alloc(struct bpf_map *map)
197 struct map_iter *iter;
199 iter = kzalloc(sizeof(*iter), GFP_KERNEL | __GFP_NOWARN);
203 iter->key = kzalloc(map->key_size, GFP_KERNEL | __GFP_NOWARN);
214 static void *map_seq_next(struct seq_file *m, void *v, loff_t *pos)
216 struct bpf_map *map = seq_file_to_map(m);
217 void *key = map_iter(m)->key;
221 if (map_iter(m)->done)
224 if (unlikely(v == SEQ_START_TOKEN))
230 if (map->ops->map_get_next_key(map, prev_key, key)) {
231 map_iter(m)->done = true;
238 static void *map_seq_start(struct seq_file *m, loff_t *pos)
240 if (map_iter(m)->done)
243 return *pos ? map_iter(m)->key : SEQ_START_TOKEN;
246 static void map_seq_stop(struct seq_file *m, void *v)
250 static int map_seq_show(struct seq_file *m, void *v)
252 struct bpf_map *map = seq_file_to_map(m);
253 void *key = map_iter(m)->key;
255 if (unlikely(v == SEQ_START_TOKEN)) {
256 seq_puts(m, "# WARNING!! The output is for debug purpose only\n");
257 seq_puts(m, "# WARNING!! The output format will change\n");
259 map->ops->map_seq_show_elem(map, key, m);
265 static const struct seq_operations bpffs_map_seq_ops = {
266 .start = map_seq_start,
267 .next = map_seq_next,
268 .show = map_seq_show,
269 .stop = map_seq_stop,
272 static int bpffs_map_open(struct inode *inode, struct file *file)
274 struct bpf_map *map = inode->i_private;
275 struct map_iter *iter;
279 iter = map_iter_alloc(map);
283 err = seq_open(file, &bpffs_map_seq_ops);
289 m = file->private_data;
295 static int bpffs_map_release(struct inode *inode, struct file *file)
297 struct seq_file *m = file->private_data;
299 map_iter_free(map_iter(m));
301 return seq_release(inode, file);
304 /* bpffs_map_fops should only implement the basic
305 * read operation for a BPF map. The purpose is to
306 * provide a simple user intuitive way to do
307 * "cat bpffs/pathto/a-pinned-map".
309 * Other operations (e.g. write, lookup...) should be realized by
310 * the userspace tools (e.g. bpftool) through the
311 * BPF_OBJ_GET_INFO_BY_FD and the map's lookup/update
314 static const struct file_operations bpffs_map_fops = {
315 .open = bpffs_map_open,
317 .release = bpffs_map_release,
320 static int bpffs_obj_open(struct inode *inode, struct file *file)
325 static const struct file_operations bpffs_obj_fops = {
326 .open = bpffs_obj_open,
329 static int bpf_mkobj_ops(struct dentry *dentry, umode_t mode, void *raw,
330 const struct inode_operations *iops,
331 const struct file_operations *fops)
333 struct inode *dir = dentry->d_parent->d_inode;
334 struct inode *inode = bpf_get_inode(dir->i_sb, dir, mode);
336 return PTR_ERR(inode);
340 inode->i_private = raw;
342 bpf_dentry_finalize(dentry, inode, dir);
346 static int bpf_mkprog(struct dentry *dentry, umode_t mode, void *arg)
348 return bpf_mkobj_ops(dentry, mode, arg, &bpf_prog_iops,
352 static int bpf_mkmap(struct dentry *dentry, umode_t mode, void *arg)
354 struct bpf_map *map = arg;
356 return bpf_mkobj_ops(dentry, mode, arg, &bpf_map_iops,
357 bpf_map_support_seq_show(map) ?
358 &bpffs_map_fops : &bpffs_obj_fops);
361 static int bpf_mklink(struct dentry *dentry, umode_t mode, void *arg)
363 struct bpf_link *link = arg;
365 return bpf_mkobj_ops(dentry, mode, arg, &bpf_link_iops,
366 bpf_link_is_iter(link) ?
367 &bpf_iter_fops : &bpffs_obj_fops);
370 static struct dentry *
371 bpf_lookup(struct inode *dir, struct dentry *dentry, unsigned flags)
373 /* Dots in names (e.g. "/sys/fs/bpf/foo.bar") are reserved for future
374 * extensions. That allows popoulate_bpffs() create special files.
376 if ((dir->i_mode & S_IALLUGO) &&
377 strchr(dentry->d_name.name, '.'))
378 return ERR_PTR(-EPERM);
380 return simple_lookup(dir, dentry, flags);
383 static int bpf_symlink(struct mnt_idmap *idmap, struct inode *dir,
384 struct dentry *dentry, const char *target)
386 char *link = kstrdup(target, GFP_USER | __GFP_NOWARN);
392 inode = bpf_get_inode(dir->i_sb, dir, S_IRWXUGO | S_IFLNK);
395 return PTR_ERR(inode);
398 inode->i_op = &simple_symlink_inode_operations;
399 inode->i_link = link;
401 bpf_dentry_finalize(dentry, inode, dir);
405 static const struct inode_operations bpf_dir_iops = {
406 .lookup = bpf_lookup,
408 .symlink = bpf_symlink,
409 .rmdir = simple_rmdir,
410 .rename = simple_rename,
412 .unlink = simple_unlink,
415 /* pin iterator link into bpffs */
416 static int bpf_iter_link_pin_kernel(struct dentry *parent,
417 const char *name, struct bpf_link *link)
419 umode_t mode = S_IFREG | S_IRUSR;
420 struct dentry *dentry;
423 inode_lock(parent->d_inode);
424 dentry = lookup_one_len(name, parent, strlen(name));
425 if (IS_ERR(dentry)) {
426 inode_unlock(parent->d_inode);
427 return PTR_ERR(dentry);
429 ret = bpf_mkobj_ops(dentry, mode, link, &bpf_link_iops,
432 inode_unlock(parent->d_inode);
436 static int bpf_obj_do_pin(int path_fd, const char __user *pathname, void *raw,
439 struct dentry *dentry;
445 dentry = user_path_create(path_fd, pathname, &path, 0);
447 return PTR_ERR(dentry);
449 dir = d_inode(path.dentry);
450 if (dir->i_op != &bpf_dir_iops) {
455 mode = S_IFREG | ((S_IRUSR | S_IWUSR) & ~current_umask());
456 ret = security_path_mknod(&path, dentry, mode, 0);
462 ret = vfs_mkobj(dentry, mode, bpf_mkprog, raw);
465 ret = vfs_mkobj(dentry, mode, bpf_mkmap, raw);
468 ret = vfs_mkobj(dentry, mode, bpf_mklink, raw);
474 done_path_create(&path, dentry);
478 int bpf_obj_pin_user(u32 ufd, int path_fd, const char __user *pathname)
484 raw = bpf_fd_probe_obj(ufd, &type);
488 ret = bpf_obj_do_pin(path_fd, pathname, raw, type);
490 bpf_any_put(raw, type);
495 static void *bpf_obj_do_get(int path_fd, const char __user *pathname,
496 enum bpf_type *type, int flags)
503 ret = user_path_at(path_fd, pathname, LOOKUP_FOLLOW, &path);
507 inode = d_backing_inode(path.dentry);
508 ret = path_permission(&path, ACC_MODE(flags));
512 ret = bpf_inode_type(inode, type);
516 raw = bpf_any_get(inode->i_private, *type);
527 int bpf_obj_get_user(int path_fd, const char __user *pathname, int flags)
529 enum bpf_type type = BPF_TYPE_UNSPEC;
534 f_flags = bpf_get_file_flag(flags);
538 raw = bpf_obj_do_get(path_fd, pathname, &type, f_flags);
542 if (type == BPF_TYPE_PROG)
543 ret = bpf_prog_new_fd(raw);
544 else if (type == BPF_TYPE_MAP)
545 ret = bpf_map_new_fd(raw, f_flags);
546 else if (type == BPF_TYPE_LINK)
547 ret = (f_flags != O_RDWR) ? -EINVAL : bpf_link_new_fd(raw);
552 bpf_any_put(raw, type);
556 static struct bpf_prog *__get_prog_inode(struct inode *inode, enum bpf_prog_type type)
558 struct bpf_prog *prog;
559 int ret = inode_permission(&nop_mnt_idmap, inode, MAY_READ);
563 if (inode->i_op == &bpf_map_iops)
564 return ERR_PTR(-EINVAL);
565 if (inode->i_op == &bpf_link_iops)
566 return ERR_PTR(-EINVAL);
567 if (inode->i_op != &bpf_prog_iops)
568 return ERR_PTR(-EACCES);
570 prog = inode->i_private;
572 ret = security_bpf_prog(prog);
576 if (!bpf_prog_get_ok(prog, &type, false))
577 return ERR_PTR(-EINVAL);
583 struct bpf_prog *bpf_prog_get_type_path(const char *name, enum bpf_prog_type type)
585 struct bpf_prog *prog;
587 int ret = kern_path(name, LOOKUP_FOLLOW, &path);
590 prog = __get_prog_inode(d_backing_inode(path.dentry), type);
596 EXPORT_SYMBOL(bpf_prog_get_type_path);
599 * Display the mount options in /proc/mounts.
601 static int bpf_show_options(struct seq_file *m, struct dentry *root)
603 umode_t mode = d_inode(root)->i_mode & S_IALLUGO & ~S_ISVTX;
605 if (mode != S_IRWXUGO)
606 seq_printf(m, ",mode=%o", mode);
610 static void bpf_free_inode(struct inode *inode)
614 if (S_ISLNK(inode->i_mode))
615 kfree(inode->i_link);
616 if (!bpf_inode_type(inode, &type))
617 bpf_any_put(inode->i_private, type);
618 free_inode_nonrcu(inode);
621 static const struct super_operations bpf_super_ops = {
622 .statfs = simple_statfs,
623 .drop_inode = generic_delete_inode,
624 .show_options = bpf_show_options,
625 .free_inode = bpf_free_inode,
632 static const struct fs_parameter_spec bpf_fs_parameters[] = {
633 fsparam_u32oct ("mode", OPT_MODE),
637 struct bpf_mount_opts {
641 static int bpf_parse_param(struct fs_context *fc, struct fs_parameter *param)
643 struct bpf_mount_opts *opts = fc->fs_private;
644 struct fs_parse_result result;
647 opt = fs_parse(fc, bpf_fs_parameters, param, &result);
649 /* We might like to report bad mount options here, but
650 * traditionally we've ignored all mount options, so we'd
651 * better continue to ignore non-existing options for bpf.
653 if (opt == -ENOPARAM) {
654 opt = vfs_parse_fs_param_source(fc, param);
655 if (opt != -ENOPARAM)
667 opts->mode = result.uint_32 & S_IALLUGO;
674 struct bpf_preload_ops *bpf_preload_ops;
675 EXPORT_SYMBOL_GPL(bpf_preload_ops);
677 static bool bpf_preload_mod_get(void)
679 /* If bpf_preload.ko wasn't loaded earlier then load it now.
680 * When bpf_preload is built into vmlinux the module's __init
681 * function will populate it.
683 if (!bpf_preload_ops) {
684 request_module("bpf_preload");
685 if (!bpf_preload_ops)
688 /* And grab the reference, so the module doesn't disappear while the
689 * kernel is interacting with the kernel module and its UMD.
691 if (!try_module_get(bpf_preload_ops->owner)) {
692 pr_err("bpf_preload module get failed.\n");
698 static void bpf_preload_mod_put(void)
701 /* now user can "rmmod bpf_preload" if necessary */
702 module_put(bpf_preload_ops->owner);
705 static DEFINE_MUTEX(bpf_preload_lock);
707 static int populate_bpffs(struct dentry *parent)
709 struct bpf_preload_info objs[BPF_PRELOAD_LINKS] = {};
712 /* grab the mutex to make sure the kernel interactions with bpf_preload
715 mutex_lock(&bpf_preload_lock);
717 /* if bpf_preload.ko wasn't built into vmlinux then load it */
718 if (!bpf_preload_mod_get())
721 err = bpf_preload_ops->preload(objs);
724 for (i = 0; i < BPF_PRELOAD_LINKS; i++) {
725 bpf_link_inc(objs[i].link);
726 err = bpf_iter_link_pin_kernel(parent,
727 objs[i].link_name, objs[i].link);
729 bpf_link_put(objs[i].link);
734 bpf_preload_mod_put();
736 mutex_unlock(&bpf_preload_lock);
740 static int bpf_fill_super(struct super_block *sb, struct fs_context *fc)
742 static const struct tree_descr bpf_rfiles[] = { { "" } };
743 struct bpf_mount_opts *opts = fc->fs_private;
747 ret = simple_fill_super(sb, BPF_FS_MAGIC, bpf_rfiles);
751 sb->s_op = &bpf_super_ops;
753 inode = sb->s_root->d_inode;
754 inode->i_op = &bpf_dir_iops;
755 inode->i_mode &= ~S_IALLUGO;
756 populate_bpffs(sb->s_root);
757 inode->i_mode |= S_ISVTX | opts->mode;
761 static int bpf_get_tree(struct fs_context *fc)
763 return get_tree_nodev(fc, bpf_fill_super);
766 static void bpf_free_fc(struct fs_context *fc)
768 kfree(fc->fs_private);
771 static const struct fs_context_operations bpf_context_ops = {
773 .parse_param = bpf_parse_param,
774 .get_tree = bpf_get_tree,
778 * Set up the filesystem mount context.
780 static int bpf_init_fs_context(struct fs_context *fc)
782 struct bpf_mount_opts *opts;
784 opts = kzalloc(sizeof(struct bpf_mount_opts), GFP_KERNEL);
788 opts->mode = S_IRWXUGO;
790 fc->fs_private = opts;
791 fc->ops = &bpf_context_ops;
795 static struct file_system_type bpf_fs_type = {
796 .owner = THIS_MODULE,
798 .init_fs_context = bpf_init_fs_context,
799 .parameters = bpf_fs_parameters,
800 .kill_sb = kill_litter_super,
803 static int __init bpf_init(void)
807 ret = sysfs_create_mount_point(fs_kobj, "bpf");
811 ret = register_filesystem(&bpf_fs_type);
813 sysfs_remove_mount_point(fs_kobj, "bpf");
817 fs_initcall(bpf_init);