Merge tag 'regmap-fix-v5.11-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git...
[linux/fpc-iii.git] / kernel / bpf / inode.c
blobdd4b7fd60ee7dbb9dea1ed09e91cc789a0a7bbc5
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Minimal file system backend for holding eBPF maps and programs,
4 * used by bpf(2) object pinning.
6 * Authors:
8 * Daniel Borkmann <daniel@iogearbox.net>
9 */
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>
16 #include <linux/fs.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"
25 enum bpf_type {
26 BPF_TYPE_UNSPEC = 0,
27 BPF_TYPE_PROG,
28 BPF_TYPE_MAP,
29 BPF_TYPE_LINK,
32 static void *bpf_any_get(void *raw, enum bpf_type type)
34 switch (type) {
35 case BPF_TYPE_PROG:
36 bpf_prog_inc(raw);
37 break;
38 case BPF_TYPE_MAP:
39 bpf_map_inc_with_uref(raw);
40 break;
41 case BPF_TYPE_LINK:
42 bpf_link_inc(raw);
43 break;
44 default:
45 WARN_ON_ONCE(1);
46 break;
49 return raw;
52 static void bpf_any_put(void *raw, enum bpf_type type)
54 switch (type) {
55 case BPF_TYPE_PROG:
56 bpf_prog_put(raw);
57 break;
58 case BPF_TYPE_MAP:
59 bpf_map_put_with_uref(raw);
60 break;
61 case BPF_TYPE_LINK:
62 bpf_link_put(raw);
63 break;
64 default:
65 WARN_ON_ONCE(1);
66 break;
70 static void *bpf_fd_probe_obj(u32 ufd, enum bpf_type *type)
72 void *raw;
74 raw = bpf_map_get_with_uref(ufd);
75 if (!IS_ERR(raw)) {
76 *type = BPF_TYPE_MAP;
77 return raw;
80 raw = bpf_prog_get(ufd);
81 if (!IS_ERR(raw)) {
82 *type = BPF_TYPE_PROG;
83 return raw;
86 raw = bpf_link_get_from_fd(ufd);
87 if (!IS_ERR(raw)) {
88 *type = BPF_TYPE_LINK;
89 return raw;
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,
103 umode_t mode)
105 struct inode *inode;
107 switch (mode & S_IFMT) {
108 case S_IFDIR:
109 case S_IFREG:
110 case S_IFLNK:
111 break;
112 default:
113 return ERR_PTR(-EINVAL);
116 inode = new_inode(sb);
117 if (!inode)
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(inode, dir, mode);
127 return inode;
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;
139 else
140 return -EACCES;
142 return 0;
145 static void bpf_dentry_finalize(struct dentry *dentry, struct inode *inode,
146 struct inode *dir)
148 d_instantiate(dentry, inode);
149 dget(dentry);
151 dir->i_mtime = current_time(dir);
152 dir->i_ctime = dir->i_mtime;
155 static int bpf_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
157 struct inode *inode;
159 inode = bpf_get_inode(dir->i_sb, dir, mode | S_IFDIR);
160 if (IS_ERR(inode))
161 return PTR_ERR(inode);
163 inode->i_op = &bpf_dir_iops;
164 inode->i_fop = &simple_dir_operations;
166 inc_nlink(inode);
167 inc_nlink(dir);
169 bpf_dentry_finalize(dentry, inode, dir);
170 return 0;
173 struct map_iter {
174 void *key;
175 bool done;
178 static struct map_iter *map_iter(struct seq_file *m)
180 return m->private;
183 static struct bpf_map *seq_file_to_map(struct seq_file *m)
185 return file_inode(m->file)->i_private;
188 static void map_iter_free(struct map_iter *iter)
190 if (iter) {
191 kfree(iter->key);
192 kfree(iter);
196 static struct map_iter *map_iter_alloc(struct bpf_map *map)
198 struct map_iter *iter;
200 iter = kzalloc(sizeof(*iter), GFP_KERNEL | __GFP_NOWARN);
201 if (!iter)
202 goto error;
204 iter->key = kzalloc(map->key_size, GFP_KERNEL | __GFP_NOWARN);
205 if (!iter->key)
206 goto error;
208 return iter;
210 error:
211 map_iter_free(iter);
212 return NULL;
215 static void *map_seq_next(struct seq_file *m, void *v, loff_t *pos)
217 struct bpf_map *map = seq_file_to_map(m);
218 void *key = map_iter(m)->key;
219 void *prev_key;
221 (*pos)++;
222 if (map_iter(m)->done)
223 return NULL;
225 if (unlikely(v == SEQ_START_TOKEN))
226 prev_key = NULL;
227 else
228 prev_key = key;
230 rcu_read_lock();
231 if (map->ops->map_get_next_key(map, prev_key, key)) {
232 map_iter(m)->done = true;
233 key = NULL;
235 rcu_read_unlock();
236 return key;
239 static void *map_seq_start(struct seq_file *m, loff_t *pos)
241 if (map_iter(m)->done)
242 return NULL;
244 return *pos ? map_iter(m)->key : SEQ_START_TOKEN;
247 static void map_seq_stop(struct seq_file *m, void *v)
251 static int map_seq_show(struct seq_file *m, void *v)
253 struct bpf_map *map = seq_file_to_map(m);
254 void *key = map_iter(m)->key;
256 if (unlikely(v == SEQ_START_TOKEN)) {
257 seq_puts(m, "# WARNING!! The output is for debug purpose only\n");
258 seq_puts(m, "# WARNING!! The output format will change\n");
259 } else {
260 map->ops->map_seq_show_elem(map, key, m);
263 return 0;
266 static const struct seq_operations bpffs_map_seq_ops = {
267 .start = map_seq_start,
268 .next = map_seq_next,
269 .show = map_seq_show,
270 .stop = map_seq_stop,
273 static int bpffs_map_open(struct inode *inode, struct file *file)
275 struct bpf_map *map = inode->i_private;
276 struct map_iter *iter;
277 struct seq_file *m;
278 int err;
280 iter = map_iter_alloc(map);
281 if (!iter)
282 return -ENOMEM;
284 err = seq_open(file, &bpffs_map_seq_ops);
285 if (err) {
286 map_iter_free(iter);
287 return err;
290 m = file->private_data;
291 m->private = iter;
293 return 0;
296 static int bpffs_map_release(struct inode *inode, struct file *file)
298 struct seq_file *m = file->private_data;
300 map_iter_free(map_iter(m));
302 return seq_release(inode, file);
305 /* bpffs_map_fops should only implement the basic
306 * read operation for a BPF map. The purpose is to
307 * provide a simple user intuitive way to do
308 * "cat bpffs/pathto/a-pinned-map".
310 * Other operations (e.g. write, lookup...) should be realized by
311 * the userspace tools (e.g. bpftool) through the
312 * BPF_OBJ_GET_INFO_BY_FD and the map's lookup/update
313 * interface.
315 static const struct file_operations bpffs_map_fops = {
316 .open = bpffs_map_open,
317 .read = seq_read,
318 .release = bpffs_map_release,
321 static int bpffs_obj_open(struct inode *inode, struct file *file)
323 return -EIO;
326 static const struct file_operations bpffs_obj_fops = {
327 .open = bpffs_obj_open,
330 static int bpf_mkobj_ops(struct dentry *dentry, umode_t mode, void *raw,
331 const struct inode_operations *iops,
332 const struct file_operations *fops)
334 struct inode *dir = dentry->d_parent->d_inode;
335 struct inode *inode = bpf_get_inode(dir->i_sb, dir, mode);
336 if (IS_ERR(inode))
337 return PTR_ERR(inode);
339 inode->i_op = iops;
340 inode->i_fop = fops;
341 inode->i_private = raw;
343 bpf_dentry_finalize(dentry, inode, dir);
344 return 0;
347 static int bpf_mkprog(struct dentry *dentry, umode_t mode, void *arg)
349 return bpf_mkobj_ops(dentry, mode, arg, &bpf_prog_iops,
350 &bpffs_obj_fops);
353 static int bpf_mkmap(struct dentry *dentry, umode_t mode, void *arg)
355 struct bpf_map *map = arg;
357 return bpf_mkobj_ops(dentry, mode, arg, &bpf_map_iops,
358 bpf_map_support_seq_show(map) ?
359 &bpffs_map_fops : &bpffs_obj_fops);
362 static int bpf_mklink(struct dentry *dentry, umode_t mode, void *arg)
364 struct bpf_link *link = arg;
366 return bpf_mkobj_ops(dentry, mode, arg, &bpf_link_iops,
367 bpf_link_is_iter(link) ?
368 &bpf_iter_fops : &bpffs_obj_fops);
371 static struct dentry *
372 bpf_lookup(struct inode *dir, struct dentry *dentry, unsigned flags)
374 /* Dots in names (e.g. "/sys/fs/bpf/foo.bar") are reserved for future
375 * extensions. That allows popoulate_bpffs() create special files.
377 if ((dir->i_mode & S_IALLUGO) &&
378 strchr(dentry->d_name.name, '.'))
379 return ERR_PTR(-EPERM);
381 return simple_lookup(dir, dentry, flags);
384 static int bpf_symlink(struct inode *dir, struct dentry *dentry,
385 const char *target)
387 char *link = kstrdup(target, GFP_USER | __GFP_NOWARN);
388 struct inode *inode;
390 if (!link)
391 return -ENOMEM;
393 inode = bpf_get_inode(dir->i_sb, dir, S_IRWXUGO | S_IFLNK);
394 if (IS_ERR(inode)) {
395 kfree(link);
396 return PTR_ERR(inode);
399 inode->i_op = &simple_symlink_inode_operations;
400 inode->i_link = link;
402 bpf_dentry_finalize(dentry, inode, dir);
403 return 0;
406 static const struct inode_operations bpf_dir_iops = {
407 .lookup = bpf_lookup,
408 .mkdir = bpf_mkdir,
409 .symlink = bpf_symlink,
410 .rmdir = simple_rmdir,
411 .rename = simple_rename,
412 .link = simple_link,
413 .unlink = simple_unlink,
416 /* pin iterator link into bpffs */
417 static int bpf_iter_link_pin_kernel(struct dentry *parent,
418 const char *name, struct bpf_link *link)
420 umode_t mode = S_IFREG | S_IRUSR;
421 struct dentry *dentry;
422 int ret;
424 inode_lock(parent->d_inode);
425 dentry = lookup_one_len(name, parent, strlen(name));
426 if (IS_ERR(dentry)) {
427 inode_unlock(parent->d_inode);
428 return PTR_ERR(dentry);
430 ret = bpf_mkobj_ops(dentry, mode, link, &bpf_link_iops,
431 &bpf_iter_fops);
432 dput(dentry);
433 inode_unlock(parent->d_inode);
434 return ret;
437 static int bpf_obj_do_pin(const char __user *pathname, void *raw,
438 enum bpf_type type)
440 struct dentry *dentry;
441 struct inode *dir;
442 struct path path;
443 umode_t mode;
444 int ret;
446 dentry = user_path_create(AT_FDCWD, pathname, &path, 0);
447 if (IS_ERR(dentry))
448 return PTR_ERR(dentry);
450 mode = S_IFREG | ((S_IRUSR | S_IWUSR) & ~current_umask());
452 ret = security_path_mknod(&path, dentry, mode, 0);
453 if (ret)
454 goto out;
456 dir = d_inode(path.dentry);
457 if (dir->i_op != &bpf_dir_iops) {
458 ret = -EPERM;
459 goto out;
462 switch (type) {
463 case BPF_TYPE_PROG:
464 ret = vfs_mkobj(dentry, mode, bpf_mkprog, raw);
465 break;
466 case BPF_TYPE_MAP:
467 ret = vfs_mkobj(dentry, mode, bpf_mkmap, raw);
468 break;
469 case BPF_TYPE_LINK:
470 ret = vfs_mkobj(dentry, mode, bpf_mklink, raw);
471 break;
472 default:
473 ret = -EPERM;
475 out:
476 done_path_create(&path, dentry);
477 return ret;
480 int bpf_obj_pin_user(u32 ufd, const char __user *pathname)
482 enum bpf_type type;
483 void *raw;
484 int ret;
486 raw = bpf_fd_probe_obj(ufd, &type);
487 if (IS_ERR(raw))
488 return PTR_ERR(raw);
490 ret = bpf_obj_do_pin(pathname, raw, type);
491 if (ret != 0)
492 bpf_any_put(raw, type);
494 return ret;
497 static void *bpf_obj_do_get(const char __user *pathname,
498 enum bpf_type *type, int flags)
500 struct inode *inode;
501 struct path path;
502 void *raw;
503 int ret;
505 ret = user_path_at(AT_FDCWD, pathname, LOOKUP_FOLLOW, &path);
506 if (ret)
507 return ERR_PTR(ret);
509 inode = d_backing_inode(path.dentry);
510 ret = inode_permission(inode, ACC_MODE(flags));
511 if (ret)
512 goto out;
514 ret = bpf_inode_type(inode, type);
515 if (ret)
516 goto out;
518 raw = bpf_any_get(inode->i_private, *type);
519 if (!IS_ERR(raw))
520 touch_atime(&path);
522 path_put(&path);
523 return raw;
524 out:
525 path_put(&path);
526 return ERR_PTR(ret);
529 int bpf_obj_get_user(const char __user *pathname, int flags)
531 enum bpf_type type = BPF_TYPE_UNSPEC;
532 int f_flags;
533 void *raw;
534 int ret;
536 f_flags = bpf_get_file_flag(flags);
537 if (f_flags < 0)
538 return f_flags;
540 raw = bpf_obj_do_get(pathname, &type, f_flags);
541 if (IS_ERR(raw))
542 return PTR_ERR(raw);
544 if (type == BPF_TYPE_PROG)
545 ret = bpf_prog_new_fd(raw);
546 else if (type == BPF_TYPE_MAP)
547 ret = bpf_map_new_fd(raw, f_flags);
548 else if (type == BPF_TYPE_LINK)
549 ret = bpf_link_new_fd(raw);
550 else
551 return -ENOENT;
553 if (ret < 0)
554 bpf_any_put(raw, type);
555 return ret;
558 static struct bpf_prog *__get_prog_inode(struct inode *inode, enum bpf_prog_type type)
560 struct bpf_prog *prog;
561 int ret = inode_permission(inode, MAY_READ);
562 if (ret)
563 return ERR_PTR(ret);
565 if (inode->i_op == &bpf_map_iops)
566 return ERR_PTR(-EINVAL);
567 if (inode->i_op == &bpf_link_iops)
568 return ERR_PTR(-EINVAL);
569 if (inode->i_op != &bpf_prog_iops)
570 return ERR_PTR(-EACCES);
572 prog = inode->i_private;
574 ret = security_bpf_prog(prog);
575 if (ret < 0)
576 return ERR_PTR(ret);
578 if (!bpf_prog_get_ok(prog, &type, false))
579 return ERR_PTR(-EINVAL);
581 bpf_prog_inc(prog);
582 return prog;
585 struct bpf_prog *bpf_prog_get_type_path(const char *name, enum bpf_prog_type type)
587 struct bpf_prog *prog;
588 struct path path;
589 int ret = kern_path(name, LOOKUP_FOLLOW, &path);
590 if (ret)
591 return ERR_PTR(ret);
592 prog = __get_prog_inode(d_backing_inode(path.dentry), type);
593 if (!IS_ERR(prog))
594 touch_atime(&path);
595 path_put(&path);
596 return prog;
598 EXPORT_SYMBOL(bpf_prog_get_type_path);
601 * Display the mount options in /proc/mounts.
603 static int bpf_show_options(struct seq_file *m, struct dentry *root)
605 umode_t mode = d_inode(root)->i_mode & S_IALLUGO & ~S_ISVTX;
607 if (mode != S_IRWXUGO)
608 seq_printf(m, ",mode=%o", mode);
609 return 0;
612 static void bpf_free_inode(struct inode *inode)
614 enum bpf_type type;
616 if (S_ISLNK(inode->i_mode))
617 kfree(inode->i_link);
618 if (!bpf_inode_type(inode, &type))
619 bpf_any_put(inode->i_private, type);
620 free_inode_nonrcu(inode);
623 static const struct super_operations bpf_super_ops = {
624 .statfs = simple_statfs,
625 .drop_inode = generic_delete_inode,
626 .show_options = bpf_show_options,
627 .free_inode = bpf_free_inode,
630 enum {
631 OPT_MODE,
634 static const struct fs_parameter_spec bpf_fs_parameters[] = {
635 fsparam_u32oct ("mode", OPT_MODE),
639 struct bpf_mount_opts {
640 umode_t mode;
643 static int bpf_parse_param(struct fs_context *fc, struct fs_parameter *param)
645 struct bpf_mount_opts *opts = fc->fs_private;
646 struct fs_parse_result result;
647 int opt;
649 opt = fs_parse(fc, bpf_fs_parameters, param, &result);
650 if (opt < 0)
651 /* We might like to report bad mount options here, but
652 * traditionally we've ignored all mount options, so we'd
653 * better continue to ignore non-existing options for bpf.
655 return opt == -ENOPARAM ? 0 : opt;
657 switch (opt) {
658 case OPT_MODE:
659 opts->mode = result.uint_32 & S_IALLUGO;
660 break;
663 return 0;
666 struct bpf_preload_ops *bpf_preload_ops;
667 EXPORT_SYMBOL_GPL(bpf_preload_ops);
669 static bool bpf_preload_mod_get(void)
671 /* If bpf_preload.ko wasn't loaded earlier then load it now.
672 * When bpf_preload is built into vmlinux the module's __init
673 * function will populate it.
675 if (!bpf_preload_ops) {
676 request_module("bpf_preload");
677 if (!bpf_preload_ops)
678 return false;
680 /* And grab the reference, so the module doesn't disappear while the
681 * kernel is interacting with the kernel module and its UMD.
683 if (!try_module_get(bpf_preload_ops->owner)) {
684 pr_err("bpf_preload module get failed.\n");
685 return false;
687 return true;
690 static void bpf_preload_mod_put(void)
692 if (bpf_preload_ops)
693 /* now user can "rmmod bpf_preload" if necessary */
694 module_put(bpf_preload_ops->owner);
697 static DEFINE_MUTEX(bpf_preload_lock);
699 static int populate_bpffs(struct dentry *parent)
701 struct bpf_preload_info objs[BPF_PRELOAD_LINKS] = {};
702 struct bpf_link *links[BPF_PRELOAD_LINKS] = {};
703 int err = 0, i;
705 /* grab the mutex to make sure the kernel interactions with bpf_preload
706 * UMD are serialized
708 mutex_lock(&bpf_preload_lock);
710 /* if bpf_preload.ko wasn't built into vmlinux then load it */
711 if (!bpf_preload_mod_get())
712 goto out;
714 if (!bpf_preload_ops->info.tgid) {
715 /* preload() will start UMD that will load BPF iterator programs */
716 err = bpf_preload_ops->preload(objs);
717 if (err)
718 goto out_put;
719 for (i = 0; i < BPF_PRELOAD_LINKS; i++) {
720 links[i] = bpf_link_by_id(objs[i].link_id);
721 if (IS_ERR(links[i])) {
722 err = PTR_ERR(links[i]);
723 goto out_put;
726 for (i = 0; i < BPF_PRELOAD_LINKS; i++) {
727 err = bpf_iter_link_pin_kernel(parent,
728 objs[i].link_name, links[i]);
729 if (err)
730 goto out_put;
731 /* do not unlink successfully pinned links even
732 * if later link fails to pin
734 links[i] = NULL;
736 /* finish() will tell UMD process to exit */
737 err = bpf_preload_ops->finish();
738 if (err)
739 goto out_put;
741 out_put:
742 bpf_preload_mod_put();
743 out:
744 mutex_unlock(&bpf_preload_lock);
745 for (i = 0; i < BPF_PRELOAD_LINKS && err; i++)
746 if (!IS_ERR_OR_NULL(links[i]))
747 bpf_link_put(links[i]);
748 return err;
751 static int bpf_fill_super(struct super_block *sb, struct fs_context *fc)
753 static const struct tree_descr bpf_rfiles[] = { { "" } };
754 struct bpf_mount_opts *opts = fc->fs_private;
755 struct inode *inode;
756 int ret;
758 ret = simple_fill_super(sb, BPF_FS_MAGIC, bpf_rfiles);
759 if (ret)
760 return ret;
762 sb->s_op = &bpf_super_ops;
764 inode = sb->s_root->d_inode;
765 inode->i_op = &bpf_dir_iops;
766 inode->i_mode &= ~S_IALLUGO;
767 populate_bpffs(sb->s_root);
768 inode->i_mode |= S_ISVTX | opts->mode;
769 return 0;
772 static int bpf_get_tree(struct fs_context *fc)
774 return get_tree_nodev(fc, bpf_fill_super);
777 static void bpf_free_fc(struct fs_context *fc)
779 kfree(fc->fs_private);
782 static const struct fs_context_operations bpf_context_ops = {
783 .free = bpf_free_fc,
784 .parse_param = bpf_parse_param,
785 .get_tree = bpf_get_tree,
789 * Set up the filesystem mount context.
791 static int bpf_init_fs_context(struct fs_context *fc)
793 struct bpf_mount_opts *opts;
795 opts = kzalloc(sizeof(struct bpf_mount_opts), GFP_KERNEL);
796 if (!opts)
797 return -ENOMEM;
799 opts->mode = S_IRWXUGO;
801 fc->fs_private = opts;
802 fc->ops = &bpf_context_ops;
803 return 0;
806 static struct file_system_type bpf_fs_type = {
807 .owner = THIS_MODULE,
808 .name = "bpf",
809 .init_fs_context = bpf_init_fs_context,
810 .parameters = bpf_fs_parameters,
811 .kill_sb = kill_litter_super,
814 static int __init bpf_init(void)
816 int ret;
818 mutex_init(&bpf_preload_lock);
820 ret = sysfs_create_mount_point(fs_kobj, "bpf");
821 if (ret)
822 return ret;
824 ret = register_filesystem(&bpf_fs_type);
825 if (ret)
826 sysfs_remove_mount_point(fs_kobj, "bpf");
828 return ret;
830 fs_initcall(bpf_init);