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 = current_time(inode);
122 inode->i_mtime = inode->i_atime;
123 inode->i_ctime = inode->i_atime;
125 inode_init_owner(&init_user_ns, inode, dir, mode);
130 static int bpf_inode_type(const struct inode *inode, enum bpf_type *type)
132 *type = BPF_TYPE_UNSPEC;
133 if (inode->i_op == &bpf_prog_iops)
134 *type = BPF_TYPE_PROG;
135 else if (inode->i_op == &bpf_map_iops)
136 *type = BPF_TYPE_MAP;
137 else if (inode->i_op == &bpf_link_iops)
138 *type = BPF_TYPE_LINK;
145 static void bpf_dentry_finalize(struct dentry *dentry, struct inode *inode,
148 d_instantiate(dentry, inode);
151 dir->i_mtime = current_time(dir);
152 dir->i_ctime = dir->i_mtime;
155 static int bpf_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
156 struct dentry *dentry, umode_t mode)
160 inode = bpf_get_inode(dir->i_sb, dir, mode | S_IFDIR);
162 return PTR_ERR(inode);
164 inode->i_op = &bpf_dir_iops;
165 inode->i_fop = &simple_dir_operations;
170 bpf_dentry_finalize(dentry, inode, dir);
179 static struct map_iter *map_iter(struct seq_file *m)
184 static struct bpf_map *seq_file_to_map(struct seq_file *m)
186 return file_inode(m->file)->i_private;
189 static void map_iter_free(struct map_iter *iter)
197 static struct map_iter *map_iter_alloc(struct bpf_map *map)
199 struct map_iter *iter;
201 iter = kzalloc(sizeof(*iter), GFP_KERNEL | __GFP_NOWARN);
205 iter->key = kzalloc(map->key_size, GFP_KERNEL | __GFP_NOWARN);
216 static void *map_seq_next(struct seq_file *m, void *v, loff_t *pos)
218 struct bpf_map *map = seq_file_to_map(m);
219 void *key = map_iter(m)->key;
223 if (map_iter(m)->done)
226 if (unlikely(v == SEQ_START_TOKEN))
232 if (map->ops->map_get_next_key(map, prev_key, key)) {
233 map_iter(m)->done = true;
240 static void *map_seq_start(struct seq_file *m, loff_t *pos)
242 if (map_iter(m)->done)
245 return *pos ? map_iter(m)->key : SEQ_START_TOKEN;
248 static void map_seq_stop(struct seq_file *m, void *v)
252 static int map_seq_show(struct seq_file *m, void *v)
254 struct bpf_map *map = seq_file_to_map(m);
255 void *key = map_iter(m)->key;
257 if (unlikely(v == SEQ_START_TOKEN)) {
258 seq_puts(m, "# WARNING!! The output is for debug purpose only\n");
259 seq_puts(m, "# WARNING!! The output format will change\n");
261 map->ops->map_seq_show_elem(map, key, m);
267 static const struct seq_operations bpffs_map_seq_ops = {
268 .start = map_seq_start,
269 .next = map_seq_next,
270 .show = map_seq_show,
271 .stop = map_seq_stop,
274 static int bpffs_map_open(struct inode *inode, struct file *file)
276 struct bpf_map *map = inode->i_private;
277 struct map_iter *iter;
281 iter = map_iter_alloc(map);
285 err = seq_open(file, &bpffs_map_seq_ops);
291 m = file->private_data;
297 static int bpffs_map_release(struct inode *inode, struct file *file)
299 struct seq_file *m = file->private_data;
301 map_iter_free(map_iter(m));
303 return seq_release(inode, file);
306 /* bpffs_map_fops should only implement the basic
307 * read operation for a BPF map. The purpose is to
308 * provide a simple user intuitive way to do
309 * "cat bpffs/pathto/a-pinned-map".
311 * Other operations (e.g. write, lookup...) should be realized by
312 * the userspace tools (e.g. bpftool) through the
313 * BPF_OBJ_GET_INFO_BY_FD and the map's lookup/update
316 static const struct file_operations bpffs_map_fops = {
317 .open = bpffs_map_open,
319 .release = bpffs_map_release,
322 static int bpffs_obj_open(struct inode *inode, struct file *file)
327 static const struct file_operations bpffs_obj_fops = {
328 .open = bpffs_obj_open,
331 static int bpf_mkobj_ops(struct dentry *dentry, umode_t mode, void *raw,
332 const struct inode_operations *iops,
333 const struct file_operations *fops)
335 struct inode *dir = dentry->d_parent->d_inode;
336 struct inode *inode = bpf_get_inode(dir->i_sb, dir, mode);
338 return PTR_ERR(inode);
342 inode->i_private = raw;
344 bpf_dentry_finalize(dentry, inode, dir);
348 static int bpf_mkprog(struct dentry *dentry, umode_t mode, void *arg)
350 return bpf_mkobj_ops(dentry, mode, arg, &bpf_prog_iops,
354 static int bpf_mkmap(struct dentry *dentry, umode_t mode, void *arg)
356 struct bpf_map *map = arg;
358 return bpf_mkobj_ops(dentry, mode, arg, &bpf_map_iops,
359 bpf_map_support_seq_show(map) ?
360 &bpffs_map_fops : &bpffs_obj_fops);
363 static int bpf_mklink(struct dentry *dentry, umode_t mode, void *arg)
365 struct bpf_link *link = arg;
367 return bpf_mkobj_ops(dentry, mode, arg, &bpf_link_iops,
368 bpf_link_is_iter(link) ?
369 &bpf_iter_fops : &bpffs_obj_fops);
372 static struct dentry *
373 bpf_lookup(struct inode *dir, struct dentry *dentry, unsigned flags)
375 /* Dots in names (e.g. "/sys/fs/bpf/foo.bar") are reserved for future
376 * extensions. That allows popoulate_bpffs() create special files.
378 if ((dir->i_mode & S_IALLUGO) &&
379 strchr(dentry->d_name.name, '.'))
380 return ERR_PTR(-EPERM);
382 return simple_lookup(dir, dentry, flags);
385 static int bpf_symlink(struct user_namespace *mnt_userns, struct inode *dir,
386 struct dentry *dentry, const char *target)
388 char *link = kstrdup(target, GFP_USER | __GFP_NOWARN);
394 inode = bpf_get_inode(dir->i_sb, dir, S_IRWXUGO | S_IFLNK);
397 return PTR_ERR(inode);
400 inode->i_op = &simple_symlink_inode_operations;
401 inode->i_link = link;
403 bpf_dentry_finalize(dentry, inode, dir);
407 static const struct inode_operations bpf_dir_iops = {
408 .lookup = bpf_lookup,
410 .symlink = bpf_symlink,
411 .rmdir = simple_rmdir,
412 .rename = simple_rename,
414 .unlink = simple_unlink,
417 /* pin iterator link into bpffs */
418 static int bpf_iter_link_pin_kernel(struct dentry *parent,
419 const char *name, struct bpf_link *link)
421 umode_t mode = S_IFREG | S_IRUSR;
422 struct dentry *dentry;
425 inode_lock(parent->d_inode);
426 dentry = lookup_one_len(name, parent, strlen(name));
427 if (IS_ERR(dentry)) {
428 inode_unlock(parent->d_inode);
429 return PTR_ERR(dentry);
431 ret = bpf_mkobj_ops(dentry, mode, link, &bpf_link_iops,
434 inode_unlock(parent->d_inode);
438 static int bpf_obj_do_pin(const char __user *pathname, void *raw,
441 struct dentry *dentry;
447 dentry = user_path_create(AT_FDCWD, pathname, &path, 0);
449 return PTR_ERR(dentry);
451 mode = S_IFREG | ((S_IRUSR | S_IWUSR) & ~current_umask());
453 ret = security_path_mknod(&path, dentry, mode, 0);
457 dir = d_inode(path.dentry);
458 if (dir->i_op != &bpf_dir_iops) {
465 ret = vfs_mkobj(dentry, mode, bpf_mkprog, raw);
468 ret = vfs_mkobj(dentry, mode, bpf_mkmap, raw);
471 ret = vfs_mkobj(dentry, mode, bpf_mklink, raw);
477 done_path_create(&path, dentry);
481 int bpf_obj_pin_user(u32 ufd, const char __user *pathname)
487 raw = bpf_fd_probe_obj(ufd, &type);
491 ret = bpf_obj_do_pin(pathname, raw, type);
493 bpf_any_put(raw, type);
498 static void *bpf_obj_do_get(const char __user *pathname,
499 enum bpf_type *type, int flags)
506 ret = user_path_at(AT_FDCWD, pathname, LOOKUP_FOLLOW, &path);
510 inode = d_backing_inode(path.dentry);
511 ret = path_permission(&path, ACC_MODE(flags));
515 ret = bpf_inode_type(inode, type);
519 raw = bpf_any_get(inode->i_private, *type);
530 int bpf_obj_get_user(const char __user *pathname, int flags)
532 enum bpf_type type = BPF_TYPE_UNSPEC;
537 f_flags = bpf_get_file_flag(flags);
541 raw = bpf_obj_do_get(pathname, &type, f_flags);
545 if (type == BPF_TYPE_PROG)
546 ret = bpf_prog_new_fd(raw);
547 else if (type == BPF_TYPE_MAP)
548 ret = bpf_map_new_fd(raw, f_flags);
549 else if (type == BPF_TYPE_LINK)
550 ret = (f_flags != O_RDWR) ? -EINVAL : bpf_link_new_fd(raw);
555 bpf_any_put(raw, type);
559 static struct bpf_prog *__get_prog_inode(struct inode *inode, enum bpf_prog_type type)
561 struct bpf_prog *prog;
562 int ret = inode_permission(&init_user_ns, inode, MAY_READ);
566 if (inode->i_op == &bpf_map_iops)
567 return ERR_PTR(-EINVAL);
568 if (inode->i_op == &bpf_link_iops)
569 return ERR_PTR(-EINVAL);
570 if (inode->i_op != &bpf_prog_iops)
571 return ERR_PTR(-EACCES);
573 prog = inode->i_private;
575 ret = security_bpf_prog(prog);
579 if (!bpf_prog_get_ok(prog, &type, false))
580 return ERR_PTR(-EINVAL);
586 struct bpf_prog *bpf_prog_get_type_path(const char *name, enum bpf_prog_type type)
588 struct bpf_prog *prog;
590 int ret = kern_path(name, LOOKUP_FOLLOW, &path);
593 prog = __get_prog_inode(d_backing_inode(path.dentry), type);
599 EXPORT_SYMBOL(bpf_prog_get_type_path);
602 * Display the mount options in /proc/mounts.
604 static int bpf_show_options(struct seq_file *m, struct dentry *root)
606 umode_t mode = d_inode(root)->i_mode & S_IALLUGO & ~S_ISVTX;
608 if (mode != S_IRWXUGO)
609 seq_printf(m, ",mode=%o", mode);
613 static void bpf_free_inode(struct inode *inode)
617 if (S_ISLNK(inode->i_mode))
618 kfree(inode->i_link);
619 if (!bpf_inode_type(inode, &type))
620 bpf_any_put(inode->i_private, type);
621 free_inode_nonrcu(inode);
624 static const struct super_operations bpf_super_ops = {
625 .statfs = simple_statfs,
626 .drop_inode = generic_delete_inode,
627 .show_options = bpf_show_options,
628 .free_inode = bpf_free_inode,
635 static const struct fs_parameter_spec bpf_fs_parameters[] = {
636 fsparam_u32oct ("mode", OPT_MODE),
640 struct bpf_mount_opts {
644 static int bpf_parse_param(struct fs_context *fc, struct fs_parameter *param)
646 struct bpf_mount_opts *opts = fc->fs_private;
647 struct fs_parse_result result;
650 opt = fs_parse(fc, bpf_fs_parameters, param, &result);
652 /* We might like to report bad mount options here, but
653 * traditionally we've ignored all mount options, so we'd
654 * better continue to ignore non-existing options for bpf.
656 if (opt == -ENOPARAM) {
657 opt = vfs_parse_fs_param_source(fc, param);
658 if (opt != -ENOPARAM)
670 opts->mode = result.uint_32 & S_IALLUGO;
677 struct bpf_preload_ops *bpf_preload_ops;
678 EXPORT_SYMBOL_GPL(bpf_preload_ops);
680 static bool bpf_preload_mod_get(void)
682 /* If bpf_preload.ko wasn't loaded earlier then load it now.
683 * When bpf_preload is built into vmlinux the module's __init
684 * function will populate it.
686 if (!bpf_preload_ops) {
687 request_module("bpf_preload");
688 if (!bpf_preload_ops)
691 /* And grab the reference, so the module doesn't disappear while the
692 * kernel is interacting with the kernel module and its UMD.
694 if (!try_module_get(bpf_preload_ops->owner)) {
695 pr_err("bpf_preload module get failed.\n");
701 static void bpf_preload_mod_put(void)
704 /* now user can "rmmod bpf_preload" if necessary */
705 module_put(bpf_preload_ops->owner);
708 static DEFINE_MUTEX(bpf_preload_lock);
710 static int populate_bpffs(struct dentry *parent)
712 struct bpf_preload_info objs[BPF_PRELOAD_LINKS] = {};
715 /* grab the mutex to make sure the kernel interactions with bpf_preload
718 mutex_lock(&bpf_preload_lock);
720 /* if bpf_preload.ko wasn't built into vmlinux then load it */
721 if (!bpf_preload_mod_get())
724 err = bpf_preload_ops->preload(objs);
727 for (i = 0; i < BPF_PRELOAD_LINKS; i++) {
728 bpf_link_inc(objs[i].link);
729 err = bpf_iter_link_pin_kernel(parent,
730 objs[i].link_name, objs[i].link);
732 bpf_link_put(objs[i].link);
737 bpf_preload_mod_put();
739 mutex_unlock(&bpf_preload_lock);
743 static int bpf_fill_super(struct super_block *sb, struct fs_context *fc)
745 static const struct tree_descr bpf_rfiles[] = { { "" } };
746 struct bpf_mount_opts *opts = fc->fs_private;
750 ret = simple_fill_super(sb, BPF_FS_MAGIC, bpf_rfiles);
754 sb->s_op = &bpf_super_ops;
756 inode = sb->s_root->d_inode;
757 inode->i_op = &bpf_dir_iops;
758 inode->i_mode &= ~S_IALLUGO;
759 populate_bpffs(sb->s_root);
760 inode->i_mode |= S_ISVTX | opts->mode;
764 static int bpf_get_tree(struct fs_context *fc)
766 return get_tree_nodev(fc, bpf_fill_super);
769 static void bpf_free_fc(struct fs_context *fc)
771 kfree(fc->fs_private);
774 static const struct fs_context_operations bpf_context_ops = {
776 .parse_param = bpf_parse_param,
777 .get_tree = bpf_get_tree,
781 * Set up the filesystem mount context.
783 static int bpf_init_fs_context(struct fs_context *fc)
785 struct bpf_mount_opts *opts;
787 opts = kzalloc(sizeof(struct bpf_mount_opts), GFP_KERNEL);
791 opts->mode = S_IRWXUGO;
793 fc->fs_private = opts;
794 fc->ops = &bpf_context_ops;
798 static struct file_system_type bpf_fs_type = {
799 .owner = THIS_MODULE,
801 .init_fs_context = bpf_init_fs_context,
802 .parameters = bpf_fs_parameters,
803 .kill_sb = kill_litter_super,
806 static int __init bpf_init(void)
810 ret = sysfs_create_mount_point(fs_kobj, "bpf");
814 ret = register_filesystem(&bpf_fs_type);
816 sysfs_remove_mount_point(fs_kobj, "bpf");
820 fs_initcall(bpf_init);