xref: /linux/kernel/bpf/inode.c (revision 8886640dade4ae2595fcdce511c8bcc716aa47d3) 
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.
5  *
6  * Authors:
7  *
8  *	Daniel Borkmann <daniel@iogearbox.net>
9  */
10 
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"
24 
25 enum bpf_type {
26 	BPF_TYPE_UNSPEC	= 0,
27 	BPF_TYPE_PROG,
28 	BPF_TYPE_MAP,
29 	BPF_TYPE_LINK,
30 };
31 
32 static void *bpf_any_get(void *raw, enum bpf_type type)
33 {
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;
47 	}
48 
49 	return raw;
50 }
51 
52 static void bpf_any_put(void *raw, enum bpf_type type)
53 {
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;
67 	}
68 }
69 
70 static void *bpf_fd_probe_obj(u32 ufd, enum bpf_type *type)
71 {
72 	void *raw;
73 
74 	raw = bpf_map_get_with_uref(ufd);
75 	if (!IS_ERR(raw)) {
76 		*type = BPF_TYPE_MAP;
77 		return raw;
78 	}
79 
80 	raw = bpf_prog_get(ufd);
81 	if (!IS_ERR(raw)) {
82 		*type = BPF_TYPE_PROG;
83 		return raw;
84 	}
85 
86 	raw = bpf_link_get_from_fd(ufd);
87 	if (!IS_ERR(raw)) {
88 		*type = BPF_TYPE_LINK;
89 		return raw;
90 	}
91 
92 	return ERR_PTR(-EINVAL);
93 }
94 
95 static const struct inode_operations bpf_dir_iops;
96 
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  = { };
100 
101 static struct inode *bpf_get_inode(struct super_block *sb,
102 				   const struct inode *dir,
103 				   umode_t mode)
104 {
105 	struct inode *inode;
106 
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);
114 	}
115 
116 	inode = new_inode(sb);
117 	if (!inode)
118 		return ERR_PTR(-ENOSPC);
119 
120 	inode->i_ino = get_next_ino();
121 	simple_inode_init_ts(inode);
122 
123 	inode_init_owner(&nop_mnt_idmap, inode, dir, mode);
124 
125 	return inode;
126 }
127 
128 static int bpf_inode_type(const struct inode *inode, enum bpf_type *type)
129 {
130 	*type = BPF_TYPE_UNSPEC;
131 	if (inode->i_op == &bpf_prog_iops)
132 		*type = BPF_TYPE_PROG;
133 	else if (inode->i_op == &bpf_map_iops)
134 		*type = BPF_TYPE_MAP;
135 	else if (inode->i_op == &bpf_link_iops)
136 		*type = BPF_TYPE_LINK;
137 	else
138 		return -EACCES;
139 
140 	return 0;
141 }
142 
143 static void bpf_dentry_finalize(struct dentry *dentry, struct inode *inode,
144 				struct inode *dir)
145 {
146 	d_instantiate(dentry, inode);
147 	dget(dentry);
148 
149 	inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
150 }
151 
152 static int bpf_mkdir(struct mnt_idmap *idmap, struct inode *dir,
153 		     struct dentry *dentry, umode_t mode)
154 {
155 	struct inode *inode;
156 
157 	inode = bpf_get_inode(dir->i_sb, dir, mode | S_IFDIR);
158 	if (IS_ERR(inode))
159 		return PTR_ERR(inode);
160 
161 	inode->i_op = &bpf_dir_iops;
162 	inode->i_fop = &simple_dir_operations;
163 
164 	inc_nlink(inode);
165 	inc_nlink(dir);
166 
167 	bpf_dentry_finalize(dentry, inode, dir);
168 	return 0;
169 }
170 
171 struct map_iter {
172 	void *key;
173 	bool done;
174 };
175 
176 static struct map_iter *map_iter(struct seq_file *m)
177 {
178 	return m->private;
179 }
180 
181 static struct bpf_map *seq_file_to_map(struct seq_file *m)
182 {
183 	return file_inode(m->file)->i_private;
184 }
185 
186 static void map_iter_free(struct map_iter *iter)
187 {
188 	if (iter) {
189 		kfree(iter->key);
190 		kfree(iter);
191 	}
192 }
193 
194 static struct map_iter *map_iter_alloc(struct bpf_map *map)
195 {
196 	struct map_iter *iter;
197 
198 	iter = kzalloc(sizeof(*iter), GFP_KERNEL | __GFP_NOWARN);
199 	if (!iter)
200 		goto error;
201 
202 	iter->key = kzalloc(map->key_size, GFP_KERNEL | __GFP_NOWARN);
203 	if (!iter->key)
204 		goto error;
205 
206 	return iter;
207 
208 error:
209 	map_iter_free(iter);
210 	return NULL;
211 }
212 
213 static void *map_seq_next(struct seq_file *m, void *v, loff_t *pos)
214 {
215 	struct bpf_map *map = seq_file_to_map(m);
216 	void *key = map_iter(m)->key;
217 	void *prev_key;
218 
219 	(*pos)++;
220 	if (map_iter(m)->done)
221 		return NULL;
222 
223 	if (unlikely(v == SEQ_START_TOKEN))
224 		prev_key = NULL;
225 	else
226 		prev_key = key;
227 
228 	rcu_read_lock();
229 	if (map->ops->map_get_next_key(map, prev_key, key)) {
230 		map_iter(m)->done = true;
231 		key = NULL;
232 	}
233 	rcu_read_unlock();
234 	return key;
235 }
236 
237 static void *map_seq_start(struct seq_file *m, loff_t *pos)
238 {
239 	if (map_iter(m)->done)
240 		return NULL;
241 
242 	return *pos ? map_iter(m)->key : SEQ_START_TOKEN;
243 }
244 
245 static void map_seq_stop(struct seq_file *m, void *v)
246 {
247 }
248 
249 static int map_seq_show(struct seq_file *m, void *v)
250 {
251 	struct bpf_map *map = seq_file_to_map(m);
252 	void *key = map_iter(m)->key;
253 
254 	if (unlikely(v == SEQ_START_TOKEN)) {
255 		seq_puts(m, "# WARNING!! The output is for debug purpose only\n");
256 		seq_puts(m, "# WARNING!! The output format will change\n");
257 	} else {
258 		map->ops->map_seq_show_elem(map, key, m);
259 	}
260 
261 	return 0;
262 }
263 
264 static const struct seq_operations bpffs_map_seq_ops = {
265 	.start	= map_seq_start,
266 	.next	= map_seq_next,
267 	.show	= map_seq_show,
268 	.stop	= map_seq_stop,
269 };
270 
271 static int bpffs_map_open(struct inode *inode, struct file *file)
272 {
273 	struct bpf_map *map = inode->i_private;
274 	struct map_iter *iter;
275 	struct seq_file *m;
276 	int err;
277 
278 	iter = map_iter_alloc(map);
279 	if (!iter)
280 		return -ENOMEM;
281 
282 	err = seq_open(file, &bpffs_map_seq_ops);
283 	if (err) {
284 		map_iter_free(iter);
285 		return err;
286 	}
287 
288 	m = file->private_data;
289 	m->private = iter;
290 
291 	return 0;
292 }
293 
294 static int bpffs_map_release(struct inode *inode, struct file *file)
295 {
296 	struct seq_file *m = file->private_data;
297 
298 	map_iter_free(map_iter(m));
299 
300 	return seq_release(inode, file);
301 }
302 
303 /* bpffs_map_fops should only implement the basic
304  * read operation for a BPF map.  The purpose is to
305  * provide a simple user intuitive way to do
306  * "cat bpffs/pathto/a-pinned-map".
307  *
308  * Other operations (e.g. write, lookup...) should be realized by
309  * the userspace tools (e.g. bpftool) through the
310  * BPF_OBJ_GET_INFO_BY_FD and the map's lookup/update
311  * interface.
312  */
313 static const struct file_operations bpffs_map_fops = {
314 	.open		= bpffs_map_open,
315 	.read		= seq_read,
316 	.release	= bpffs_map_release,
317 };
318 
319 static int bpffs_obj_open(struct inode *inode, struct file *file)
320 {
321 	return -EIO;
322 }
323 
324 static const struct file_operations bpffs_obj_fops = {
325 	.open		= bpffs_obj_open,
326 };
327 
328 static int bpf_mkobj_ops(struct dentry *dentry, umode_t mode, void *raw,
329 			 const struct inode_operations *iops,
330 			 const struct file_operations *fops)
331 {
332 	struct inode *dir = dentry->d_parent->d_inode;
333 	struct inode *inode = bpf_get_inode(dir->i_sb, dir, mode);
334 	if (IS_ERR(inode))
335 		return PTR_ERR(inode);
336 
337 	inode->i_op = iops;
338 	inode->i_fop = fops;
339 	inode->i_private = raw;
340 
341 	bpf_dentry_finalize(dentry, inode, dir);
342 	return 0;
343 }
344 
345 static int bpf_mkprog(struct dentry *dentry, umode_t mode, void *arg)
346 {
347 	return bpf_mkobj_ops(dentry, mode, arg, &bpf_prog_iops,
348 			     &bpffs_obj_fops);
349 }
350 
351 static int bpf_mkmap(struct dentry *dentry, umode_t mode, void *arg)
352 {
353 	struct bpf_map *map = arg;
354 
355 	return bpf_mkobj_ops(dentry, mode, arg, &bpf_map_iops,
356 			     bpf_map_support_seq_show(map) ?
357 			     &bpffs_map_fops : &bpffs_obj_fops);
358 }
359 
360 static int bpf_mklink(struct dentry *dentry, umode_t mode, void *arg)
361 {
362 	struct bpf_link *link = arg;
363 
364 	return bpf_mkobj_ops(dentry, mode, arg, &bpf_link_iops,
365 			     bpf_link_is_iter(link) ?
366 			     &bpf_iter_fops : &bpffs_obj_fops);
367 }
368 
369 static struct dentry *
370 bpf_lookup(struct inode *dir, struct dentry *dentry, unsigned flags)
371 {
372 	/* Dots in names (e.g. "/sys/fs/bpf/foo.bar") are reserved for future
373 	 * extensions. That allows popoulate_bpffs() create special files.
374 	 */
375 	if ((dir->i_mode & S_IALLUGO) &&
376 	    strchr(dentry->d_name.name, '.'))
377 		return ERR_PTR(-EPERM);
378 
379 	return simple_lookup(dir, dentry, flags);
380 }
381 
382 static int bpf_symlink(struct mnt_idmap *idmap, struct inode *dir,
383 		       struct dentry *dentry, const char *target)
384 {
385 	char *link = kstrdup(target, GFP_USER | __GFP_NOWARN);
386 	struct inode *inode;
387 
388 	if (!link)
389 		return -ENOMEM;
390 
391 	inode = bpf_get_inode(dir->i_sb, dir, S_IRWXUGO | S_IFLNK);
392 	if (IS_ERR(inode)) {
393 		kfree(link);
394 		return PTR_ERR(inode);
395 	}
396 
397 	inode->i_op = &simple_symlink_inode_operations;
398 	inode->i_link = link;
399 
400 	bpf_dentry_finalize(dentry, inode, dir);
401 	return 0;
402 }
403 
404 static const struct inode_operations bpf_dir_iops = {
405 	.lookup		= bpf_lookup,
406 	.mkdir		= bpf_mkdir,
407 	.symlink	= bpf_symlink,
408 	.rmdir		= simple_rmdir,
409 	.rename		= simple_rename,
410 	.link		= simple_link,
411 	.unlink		= simple_unlink,
412 };
413 
414 /* pin iterator link into bpffs */
415 static int bpf_iter_link_pin_kernel(struct dentry *parent,
416 				    const char *name, struct bpf_link *link)
417 {
418 	umode_t mode = S_IFREG | S_IRUSR;
419 	struct dentry *dentry;
420 	int ret;
421 
422 	inode_lock(parent->d_inode);
423 	dentry = lookup_one_len(name, parent, strlen(name));
424 	if (IS_ERR(dentry)) {
425 		inode_unlock(parent->d_inode);
426 		return PTR_ERR(dentry);
427 	}
428 	ret = bpf_mkobj_ops(dentry, mode, link, &bpf_link_iops,
429 			    &bpf_iter_fops);
430 	dput(dentry);
431 	inode_unlock(parent->d_inode);
432 	return ret;
433 }
434 
435 static int bpf_obj_do_pin(int path_fd, const char __user *pathname, void *raw,
436 			  enum bpf_type type)
437 {
438 	struct dentry *dentry;
439 	struct inode *dir;
440 	struct path path;
441 	umode_t mode;
442 	int ret;
443 
444 	dentry = user_path_create(path_fd, pathname, &path, 0);
445 	if (IS_ERR(dentry))
446 		return PTR_ERR(dentry);
447 
448 	dir = d_inode(path.dentry);
449 	if (dir->i_op != &bpf_dir_iops) {
450 		ret = -EPERM;
451 		goto out;
452 	}
453 
454 	mode = S_IFREG | ((S_IRUSR | S_IWUSR) & ~current_umask());
455 	ret = security_path_mknod(&path, dentry, mode, 0);
456 	if (ret)
457 		goto out;
458 
459 	switch (type) {
460 	case BPF_TYPE_PROG:
461 		ret = vfs_mkobj(dentry, mode, bpf_mkprog, raw);
462 		break;
463 	case BPF_TYPE_MAP:
464 		ret = vfs_mkobj(dentry, mode, bpf_mkmap, raw);
465 		break;
466 	case BPF_TYPE_LINK:
467 		ret = vfs_mkobj(dentry, mode, bpf_mklink, raw);
468 		break;
469 	default:
470 		ret = -EPERM;
471 	}
472 out:
473 	done_path_create(&path, dentry);
474 	return ret;
475 }
476 
477 int bpf_obj_pin_user(u32 ufd, int path_fd, const char __user *pathname)
478 {
479 	enum bpf_type type;
480 	void *raw;
481 	int ret;
482 
483 	raw = bpf_fd_probe_obj(ufd, &type);
484 	if (IS_ERR(raw))
485 		return PTR_ERR(raw);
486 
487 	ret = bpf_obj_do_pin(path_fd, pathname, raw, type);
488 	if (ret != 0)
489 		bpf_any_put(raw, type);
490 
491 	return ret;
492 }
493 
494 static void *bpf_obj_do_get(int path_fd, const char __user *pathname,
495 			    enum bpf_type *type, int flags)
496 {
497 	struct inode *inode;
498 	struct path path;
499 	void *raw;
500 	int ret;
501 
502 	ret = user_path_at(path_fd, pathname, LOOKUP_FOLLOW, &path);
503 	if (ret)
504 		return ERR_PTR(ret);
505 
506 	inode = d_backing_inode(path.dentry);
507 	ret = path_permission(&path, ACC_MODE(flags));
508 	if (ret)
509 		goto out;
510 
511 	ret = bpf_inode_type(inode, type);
512 	if (ret)
513 		goto out;
514 
515 	raw = bpf_any_get(inode->i_private, *type);
516 	if (!IS_ERR(raw))
517 		touch_atime(&path);
518 
519 	path_put(&path);
520 	return raw;
521 out:
522 	path_put(&path);
523 	return ERR_PTR(ret);
524 }
525 
526 int bpf_obj_get_user(int path_fd, const char __user *pathname, int flags)
527 {
528 	enum bpf_type type = BPF_TYPE_UNSPEC;
529 	int f_flags;
530 	void *raw;
531 	int ret;
532 
533 	f_flags = bpf_get_file_flag(flags);
534 	if (f_flags < 0)
535 		return f_flags;
536 
537 	raw = bpf_obj_do_get(path_fd, pathname, &type, f_flags);
538 	if (IS_ERR(raw))
539 		return PTR_ERR(raw);
540 
541 	if (type == BPF_TYPE_PROG)
542 		ret = bpf_prog_new_fd(raw);
543 	else if (type == BPF_TYPE_MAP)
544 		ret = bpf_map_new_fd(raw, f_flags);
545 	else if (type == BPF_TYPE_LINK)
546 		ret = (f_flags != O_RDWR) ? -EINVAL : bpf_link_new_fd(raw);
547 	else
548 		return -ENOENT;
549 
550 	if (ret < 0)
551 		bpf_any_put(raw, type);
552 	return ret;
553 }
554 
555 static struct bpf_prog *__get_prog_inode(struct inode *inode, enum bpf_prog_type type)
556 {
557 	struct bpf_prog *prog;
558 	int ret = inode_permission(&nop_mnt_idmap, inode, MAY_READ);
559 	if (ret)
560 		return ERR_PTR(ret);
561 
562 	if (inode->i_op == &bpf_map_iops)
563 		return ERR_PTR(-EINVAL);
564 	if (inode->i_op == &bpf_link_iops)
565 		return ERR_PTR(-EINVAL);
566 	if (inode->i_op != &bpf_prog_iops)
567 		return ERR_PTR(-EACCES);
568 
569 	prog = inode->i_private;
570 
571 	ret = security_bpf_prog(prog);
572 	if (ret < 0)
573 		return ERR_PTR(ret);
574 
575 	if (!bpf_prog_get_ok(prog, &type, false))
576 		return ERR_PTR(-EINVAL);
577 
578 	bpf_prog_inc(prog);
579 	return prog;
580 }
581 
582 struct bpf_prog *bpf_prog_get_type_path(const char *name, enum bpf_prog_type type)
583 {
584 	struct bpf_prog *prog;
585 	struct path path;
586 	int ret = kern_path(name, LOOKUP_FOLLOW, &path);
587 	if (ret)
588 		return ERR_PTR(ret);
589 	prog = __get_prog_inode(d_backing_inode(path.dentry), type);
590 	if (!IS_ERR(prog))
591 		touch_atime(&path);
592 	path_put(&path);
593 	return prog;
594 }
595 EXPORT_SYMBOL(bpf_prog_get_type_path);
596 
597 /*
598  * Display the mount options in /proc/mounts.
599  */
600 static int bpf_show_options(struct seq_file *m, struct dentry *root)
601 {
602 	struct inode *inode = d_inode(root);
603 	umode_t mode = inode->i_mode & S_IALLUGO & ~S_ISVTX;
604 
605 	if (!uid_eq(inode->i_uid, GLOBAL_ROOT_UID))
606 		seq_printf(m, ",uid=%u",
607 			   from_kuid_munged(&init_user_ns, inode->i_uid));
608 	if (!gid_eq(inode->i_gid, GLOBAL_ROOT_GID))
609 		seq_printf(m, ",gid=%u",
610 			   from_kgid_munged(&init_user_ns, inode->i_gid));
611 	if (mode != S_IRWXUGO)
612 		seq_printf(m, ",mode=%o", mode);
613 	return 0;
614 }
615 
616 static void bpf_free_inode(struct inode *inode)
617 {
618 	enum bpf_type type;
619 
620 	if (S_ISLNK(inode->i_mode))
621 		kfree(inode->i_link);
622 	if (!bpf_inode_type(inode, &type))
623 		bpf_any_put(inode->i_private, type);
624 	free_inode_nonrcu(inode);
625 }
626 
627 static const struct super_operations bpf_super_ops = {
628 	.statfs		= simple_statfs,
629 	.drop_inode	= generic_delete_inode,
630 	.show_options	= bpf_show_options,
631 	.free_inode	= bpf_free_inode,
632 };
633 
634 enum {
635 	OPT_UID,
636 	OPT_GID,
637 	OPT_MODE,
638 };
639 
640 static const struct fs_parameter_spec bpf_fs_parameters[] = {
641 	fsparam_u32	("uid",				OPT_UID),
642 	fsparam_u32	("gid",				OPT_GID),
643 	fsparam_u32oct	("mode",			OPT_MODE),
644 	{}
645 };
646 
647 struct bpf_mount_opts {
648 	kuid_t uid;
649 	kgid_t gid;
650 	umode_t mode;
651 };
652 
653 static int bpf_parse_param(struct fs_context *fc, struct fs_parameter *param)
654 {
655 	struct bpf_mount_opts *opts = fc->fs_private;
656 	struct fs_parse_result result;
657 	kuid_t uid;
658 	kgid_t gid;
659 	int opt;
660 
661 	opt = fs_parse(fc, bpf_fs_parameters, param, &result);
662 	if (opt < 0) {
663 		/* We might like to report bad mount options here, but
664 		 * traditionally we've ignored all mount options, so we'd
665 		 * better continue to ignore non-existing options for bpf.
666 		 */
667 		if (opt == -ENOPARAM) {
668 			opt = vfs_parse_fs_param_source(fc, param);
669 			if (opt != -ENOPARAM)
670 				return opt;
671 
672 			return 0;
673 		}
674 
675 		if (opt < 0)
676 			return opt;
677 	}
678 
679 	switch (opt) {
680 	case OPT_UID:
681 		uid = make_kuid(current_user_ns(), result.uint_32);
682 		if (!uid_valid(uid))
683 			goto bad_value;
684 
685 		/*
686 		 * The requested uid must be representable in the
687 		 * filesystem's idmapping.
688 		 */
689 		if (!kuid_has_mapping(fc->user_ns, uid))
690 			goto bad_value;
691 
692 		opts->uid = uid;
693 		break;
694 	case OPT_GID:
695 		gid = make_kgid(current_user_ns(), result.uint_32);
696 		if (!gid_valid(gid))
697 			goto bad_value;
698 
699 		/*
700 		 * The requested gid must be representable in the
701 		 * filesystem's idmapping.
702 		 */
703 		if (!kgid_has_mapping(fc->user_ns, gid))
704 			goto bad_value;
705 
706 		opts->gid = gid;
707 		break;
708 	case OPT_MODE:
709 		opts->mode = result.uint_32 & S_IALLUGO;
710 		break;
711 	}
712 
713 	return 0;
714 bad_value:
715 	return invalfc(fc, "Bad value for '%s'", param->key);
716 }
717 
718 struct bpf_preload_ops *bpf_preload_ops;
719 EXPORT_SYMBOL_GPL(bpf_preload_ops);
720 
721 static bool bpf_preload_mod_get(void)
722 {
723 	/* If bpf_preload.ko wasn't loaded earlier then load it now.
724 	 * When bpf_preload is built into vmlinux the module's __init
725 	 * function will populate it.
726 	 */
727 	if (!bpf_preload_ops) {
728 		request_module("bpf_preload");
729 		if (!bpf_preload_ops)
730 			return false;
731 	}
732 	/* And grab the reference, so the module doesn't disappear while the
733 	 * kernel is interacting with the kernel module and its UMD.
734 	 */
735 	if (!try_module_get(bpf_preload_ops->owner)) {
736 		pr_err("bpf_preload module get failed.\n");
737 		return false;
738 	}
739 	return true;
740 }
741 
742 static void bpf_preload_mod_put(void)
743 {
744 	if (bpf_preload_ops)
745 		/* now user can "rmmod bpf_preload" if necessary */
746 		module_put(bpf_preload_ops->owner);
747 }
748 
749 static DEFINE_MUTEX(bpf_preload_lock);
750 
751 static int populate_bpffs(struct dentry *parent)
752 {
753 	struct bpf_preload_info objs[BPF_PRELOAD_LINKS] = {};
754 	int err = 0, i;
755 
756 	/* grab the mutex to make sure the kernel interactions with bpf_preload
757 	 * are serialized
758 	 */
759 	mutex_lock(&bpf_preload_lock);
760 
761 	/* if bpf_preload.ko wasn't built into vmlinux then load it */
762 	if (!bpf_preload_mod_get())
763 		goto out;
764 
765 	err = bpf_preload_ops->preload(objs);
766 	if (err)
767 		goto out_put;
768 	for (i = 0; i < BPF_PRELOAD_LINKS; i++) {
769 		bpf_link_inc(objs[i].link);
770 		err = bpf_iter_link_pin_kernel(parent,
771 					       objs[i].link_name, objs[i].link);
772 		if (err) {
773 			bpf_link_put(objs[i].link);
774 			goto out_put;
775 		}
776 	}
777 out_put:
778 	bpf_preload_mod_put();
779 out:
780 	mutex_unlock(&bpf_preload_lock);
781 	return err;
782 }
783 
784 static int bpf_fill_super(struct super_block *sb, struct fs_context *fc)
785 {
786 	static const struct tree_descr bpf_rfiles[] = { { "" } };
787 	struct bpf_mount_opts *opts = fc->fs_private;
788 	struct inode *inode;
789 	int ret;
790 
791 	ret = simple_fill_super(sb, BPF_FS_MAGIC, bpf_rfiles);
792 	if (ret)
793 		return ret;
794 
795 	sb->s_op = &bpf_super_ops;
796 
797 	inode = sb->s_root->d_inode;
798 	inode->i_uid = opts->uid;
799 	inode->i_gid = opts->gid;
800 	inode->i_op = &bpf_dir_iops;
801 	inode->i_mode &= ~S_IALLUGO;
802 	populate_bpffs(sb->s_root);
803 	inode->i_mode |= S_ISVTX | opts->mode;
804 	return 0;
805 }
806 
807 static int bpf_get_tree(struct fs_context *fc)
808 {
809 	return get_tree_nodev(fc, bpf_fill_super);
810 }
811 
812 static void bpf_free_fc(struct fs_context *fc)
813 {
814 	kfree(fc->fs_private);
815 }
816 
817 static const struct fs_context_operations bpf_context_ops = {
818 	.free		= bpf_free_fc,
819 	.parse_param	= bpf_parse_param,
820 	.get_tree	= bpf_get_tree,
821 };
822 
823 /*
824  * Set up the filesystem mount context.
825  */
826 static int bpf_init_fs_context(struct fs_context *fc)
827 {
828 	struct bpf_mount_opts *opts;
829 
830 	opts = kzalloc(sizeof(struct bpf_mount_opts), GFP_KERNEL);
831 	if (!opts)
832 		return -ENOMEM;
833 
834 	opts->mode = S_IRWXUGO;
835 	opts->uid = current_fsuid();
836 	opts->gid = current_fsgid();
837 
838 	fc->fs_private = opts;
839 	fc->ops = &bpf_context_ops;
840 	return 0;
841 }
842 
843 static struct file_system_type bpf_fs_type = {
844 	.owner		= THIS_MODULE,
845 	.name		= "bpf",
846 	.init_fs_context = bpf_init_fs_context,
847 	.parameters	= bpf_fs_parameters,
848 	.kill_sb	= kill_litter_super,
849 };
850 
851 static int __init bpf_init(void)
852 {
853 	int ret;
854 
855 	ret = sysfs_create_mount_point(fs_kobj, "bpf");
856 	if (ret)
857 		return ret;
858 
859 	ret = register_filesystem(&bpf_fs_type);
860 	if (ret)
861 		sysfs_remove_mount_point(fs_kobj, "bpf");
862 
863 	return ret;
864 }
865 fs_initcall(bpf_init);
866