xref: /linux-6.15/kernel/bpf/inode.c (revision 6b4a64ba) 
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 <[email protected]>
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 <linux/kstrtox.h>
24 #include "preload/bpf_preload.h"
25 
26 enum bpf_type {
27 	BPF_TYPE_UNSPEC	= 0,
28 	BPF_TYPE_PROG,
29 	BPF_TYPE_MAP,
30 	BPF_TYPE_LINK,
31 };
32 
33 static void *bpf_any_get(void *raw, enum bpf_type type)
34 {
35 	switch (type) {
36 	case BPF_TYPE_PROG:
37 		bpf_prog_inc(raw);
38 		break;
39 	case BPF_TYPE_MAP:
40 		bpf_map_inc_with_uref(raw);
41 		break;
42 	case BPF_TYPE_LINK:
43 		bpf_link_inc(raw);
44 		break;
45 	default:
46 		WARN_ON_ONCE(1);
47 		break;
48 	}
49 
50 	return raw;
51 }
52 
53 static void bpf_any_put(void *raw, enum bpf_type type)
54 {
55 	switch (type) {
56 	case BPF_TYPE_PROG:
57 		bpf_prog_put(raw);
58 		break;
59 	case BPF_TYPE_MAP:
60 		bpf_map_put_with_uref(raw);
61 		break;
62 	case BPF_TYPE_LINK:
63 		bpf_link_put(raw);
64 		break;
65 	default:
66 		WARN_ON_ONCE(1);
67 		break;
68 	}
69 }
70 
71 static void *bpf_fd_probe_obj(u32 ufd, enum bpf_type *type)
72 {
73 	void *raw;
74 
75 	raw = bpf_map_get_with_uref(ufd);
76 	if (!IS_ERR(raw)) {
77 		*type = BPF_TYPE_MAP;
78 		return raw;
79 	}
80 
81 	raw = bpf_prog_get(ufd);
82 	if (!IS_ERR(raw)) {
83 		*type = BPF_TYPE_PROG;
84 		return raw;
85 	}
86 
87 	raw = bpf_link_get_from_fd(ufd);
88 	if (!IS_ERR(raw)) {
89 		*type = BPF_TYPE_LINK;
90 		return raw;
91 	}
92 
93 	return ERR_PTR(-EINVAL);
94 }
95 
96 static const struct inode_operations bpf_dir_iops;
97 
98 static const struct inode_operations bpf_prog_iops = { };
99 static const struct inode_operations bpf_map_iops  = { };
100 static const struct inode_operations bpf_link_iops  = { };
101 
102 struct inode *bpf_get_inode(struct super_block *sb,
103 			    const struct inode *dir,
104 			    umode_t mode)
105 {
106 	struct inode *inode;
107 
108 	switch (mode & S_IFMT) {
109 	case S_IFDIR:
110 	case S_IFREG:
111 	case S_IFLNK:
112 		break;
113 	default:
114 		return ERR_PTR(-EINVAL);
115 	}
116 
117 	inode = new_inode(sb);
118 	if (!inode)
119 		return ERR_PTR(-ENOSPC);
120 
121 	inode->i_ino = get_next_ino();
122 	simple_inode_init_ts(inode);
123 
124 	inode_init_owner(&nop_mnt_idmap, inode, dir, mode);
125 
126 	return inode;
127 }
128 
129 static int bpf_inode_type(const struct inode *inode, enum bpf_type *type)
130 {
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;
138 	else
139 		return -EACCES;
140 
141 	return 0;
142 }
143 
144 static void bpf_dentry_finalize(struct dentry *dentry, struct inode *inode,
145 				struct inode *dir)
146 {
147 	d_instantiate(dentry, inode);
148 	dget(dentry);
149 
150 	inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
151 }
152 
153 static int bpf_mkdir(struct mnt_idmap *idmap, struct inode *dir,
154 		     struct dentry *dentry, umode_t mode)
155 {
156 	struct inode *inode;
157 
158 	inode = bpf_get_inode(dir->i_sb, dir, mode | S_IFDIR);
159 	if (IS_ERR(inode))
160 		return PTR_ERR(inode);
161 
162 	inode->i_op = &bpf_dir_iops;
163 	inode->i_fop = &simple_dir_operations;
164 
165 	inc_nlink(inode);
166 	inc_nlink(dir);
167 
168 	bpf_dentry_finalize(dentry, inode, dir);
169 	return 0;
170 }
171 
172 struct map_iter {
173 	void *key;
174 	bool done;
175 };
176 
177 static struct map_iter *map_iter(struct seq_file *m)
178 {
179 	return m->private;
180 }
181 
182 static struct bpf_map *seq_file_to_map(struct seq_file *m)
183 {
184 	return file_inode(m->file)->i_private;
185 }
186 
187 static void map_iter_free(struct map_iter *iter)
188 {
189 	if (iter) {
190 		kfree(iter->key);
191 		kfree(iter);
192 	}
193 }
194 
195 static struct map_iter *map_iter_alloc(struct bpf_map *map)
196 {
197 	struct map_iter *iter;
198 
199 	iter = kzalloc(sizeof(*iter), GFP_KERNEL | __GFP_NOWARN);
200 	if (!iter)
201 		goto error;
202 
203 	iter->key = kzalloc(map->key_size, GFP_KERNEL | __GFP_NOWARN);
204 	if (!iter->key)
205 		goto error;
206 
207 	return iter;
208 
209 error:
210 	map_iter_free(iter);
211 	return NULL;
212 }
213 
214 static void *map_seq_next(struct seq_file *m, void *v, loff_t *pos)
215 {
216 	struct bpf_map *map = seq_file_to_map(m);
217 	void *key = map_iter(m)->key;
218 	void *prev_key;
219 
220 	(*pos)++;
221 	if (map_iter(m)->done)
222 		return NULL;
223 
224 	if (unlikely(v == SEQ_START_TOKEN))
225 		prev_key = NULL;
226 	else
227 		prev_key = key;
228 
229 	rcu_read_lock();
230 	if (map->ops->map_get_next_key(map, prev_key, key)) {
231 		map_iter(m)->done = true;
232 		key = NULL;
233 	}
234 	rcu_read_unlock();
235 	return key;
236 }
237 
238 static void *map_seq_start(struct seq_file *m, loff_t *pos)
239 {
240 	if (map_iter(m)->done)
241 		return NULL;
242 
243 	return *pos ? map_iter(m)->key : SEQ_START_TOKEN;
244 }
245 
246 static void map_seq_stop(struct seq_file *m, void *v)
247 {
248 }
249 
250 static int map_seq_show(struct seq_file *m, void *v)
251 {
252 	struct bpf_map *map = seq_file_to_map(m);
253 	void *key = map_iter(m)->key;
254 
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");
258 	} else {
259 		map->ops->map_seq_show_elem(map, key, m);
260 	}
261 
262 	return 0;
263 }
264 
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,
270 };
271 
272 static int bpffs_map_open(struct inode *inode, struct file *file)
273 {
274 	struct bpf_map *map = inode->i_private;
275 	struct map_iter *iter;
276 	struct seq_file *m;
277 	int err;
278 
279 	iter = map_iter_alloc(map);
280 	if (!iter)
281 		return -ENOMEM;
282 
283 	err = seq_open(file, &bpffs_map_seq_ops);
284 	if (err) {
285 		map_iter_free(iter);
286 		return err;
287 	}
288 
289 	m = file->private_data;
290 	m->private = iter;
291 
292 	return 0;
293 }
294 
295 static int bpffs_map_release(struct inode *inode, struct file *file)
296 {
297 	struct seq_file *m = file->private_data;
298 
299 	map_iter_free(map_iter(m));
300 
301 	return seq_release(inode, file);
302 }
303 
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".
308  *
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
312  * interface.
313  */
314 static const struct file_operations bpffs_map_fops = {
315 	.open		= bpffs_map_open,
316 	.read		= seq_read,
317 	.release	= bpffs_map_release,
318 };
319 
320 static int bpffs_obj_open(struct inode *inode, struct file *file)
321 {
322 	return -EIO;
323 }
324 
325 static const struct file_operations bpffs_obj_fops = {
326 	.open		= bpffs_obj_open,
327 };
328 
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)
332 {
333 	struct inode *dir = dentry->d_parent->d_inode;
334 	struct inode *inode = bpf_get_inode(dir->i_sb, dir, mode);
335 	if (IS_ERR(inode))
336 		return PTR_ERR(inode);
337 
338 	inode->i_op = iops;
339 	inode->i_fop = fops;
340 	inode->i_private = raw;
341 
342 	bpf_dentry_finalize(dentry, inode, dir);
343 	return 0;
344 }
345 
346 static int bpf_mkprog(struct dentry *dentry, umode_t mode, void *arg)
347 {
348 	return bpf_mkobj_ops(dentry, mode, arg, &bpf_prog_iops,
349 			     &bpffs_obj_fops);
350 }
351 
352 static int bpf_mkmap(struct dentry *dentry, umode_t mode, void *arg)
353 {
354 	struct bpf_map *map = arg;
355 
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);
359 }
360 
361 static int bpf_mklink(struct dentry *dentry, umode_t mode, void *arg)
362 {
363 	struct bpf_link *link = arg;
364 
365 	return bpf_mkobj_ops(dentry, mode, arg, &bpf_link_iops,
366 			     bpf_link_is_iter(link) ?
367 			     &bpf_iter_fops : &bpffs_obj_fops);
368 }
369 
370 static struct dentry *
371 bpf_lookup(struct inode *dir, struct dentry *dentry, unsigned flags)
372 {
373 	/* Dots in names (e.g. "/sys/fs/bpf/foo.bar") are reserved for future
374 	 * extensions. That allows popoulate_bpffs() create special files.
375 	 */
376 	if ((dir->i_mode & S_IALLUGO) &&
377 	    strchr(dentry->d_name.name, '.'))
378 		return ERR_PTR(-EPERM);
379 
380 	return simple_lookup(dir, dentry, flags);
381 }
382 
383 static int bpf_symlink(struct mnt_idmap *idmap, struct inode *dir,
384 		       struct dentry *dentry, const char *target)
385 {
386 	char *link = kstrdup(target, GFP_USER | __GFP_NOWARN);
387 	struct inode *inode;
388 
389 	if (!link)
390 		return -ENOMEM;
391 
392 	inode = bpf_get_inode(dir->i_sb, dir, S_IRWXUGO | S_IFLNK);
393 	if (IS_ERR(inode)) {
394 		kfree(link);
395 		return PTR_ERR(inode);
396 	}
397 
398 	inode->i_op = &simple_symlink_inode_operations;
399 	inode->i_link = link;
400 
401 	bpf_dentry_finalize(dentry, inode, dir);
402 	return 0;
403 }
404 
405 static const struct inode_operations bpf_dir_iops = {
406 	.lookup		= bpf_lookup,
407 	.mkdir		= bpf_mkdir,
408 	.symlink	= bpf_symlink,
409 	.rmdir		= simple_rmdir,
410 	.rename		= simple_rename,
411 	.link		= simple_link,
412 	.unlink		= simple_unlink,
413 };
414 
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)
418 {
419 	umode_t mode = S_IFREG | S_IRUSR;
420 	struct dentry *dentry;
421 	int ret;
422 
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);
428 	}
429 	ret = bpf_mkobj_ops(dentry, mode, link, &bpf_link_iops,
430 			    &bpf_iter_fops);
431 	dput(dentry);
432 	inode_unlock(parent->d_inode);
433 	return ret;
434 }
435 
436 static int bpf_obj_do_pin(int path_fd, const char __user *pathname, void *raw,
437 			  enum bpf_type type)
438 {
439 	struct dentry *dentry;
440 	struct inode *dir;
441 	struct path path;
442 	umode_t mode;
443 	int ret;
444 
445 	dentry = user_path_create(path_fd, pathname, &path, 0);
446 	if (IS_ERR(dentry))
447 		return PTR_ERR(dentry);
448 
449 	dir = d_inode(path.dentry);
450 	if (dir->i_op != &bpf_dir_iops) {
451 		ret = -EPERM;
452 		goto out;
453 	}
454 
455 	mode = S_IFREG | ((S_IRUSR | S_IWUSR) & ~current_umask());
456 	ret = security_path_mknod(&path, dentry, mode, 0);
457 	if (ret)
458 		goto out;
459 
460 	switch (type) {
461 	case BPF_TYPE_PROG:
462 		ret = vfs_mkobj(dentry, mode, bpf_mkprog, raw);
463 		break;
464 	case BPF_TYPE_MAP:
465 		ret = vfs_mkobj(dentry, mode, bpf_mkmap, raw);
466 		break;
467 	case BPF_TYPE_LINK:
468 		ret = vfs_mkobj(dentry, mode, bpf_mklink, raw);
469 		break;
470 	default:
471 		ret = -EPERM;
472 	}
473 out:
474 	done_path_create(&path, dentry);
475 	return ret;
476 }
477 
478 int bpf_obj_pin_user(u32 ufd, int path_fd, const char __user *pathname)
479 {
480 	enum bpf_type type;
481 	void *raw;
482 	int ret;
483 
484 	raw = bpf_fd_probe_obj(ufd, &type);
485 	if (IS_ERR(raw))
486 		return PTR_ERR(raw);
487 
488 	ret = bpf_obj_do_pin(path_fd, pathname, raw, type);
489 	if (ret != 0)
490 		bpf_any_put(raw, type);
491 
492 	return ret;
493 }
494 
495 static void *bpf_obj_do_get(int path_fd, const char __user *pathname,
496 			    enum bpf_type *type, int flags)
497 {
498 	struct inode *inode;
499 	struct path path;
500 	void *raw;
501 	int ret;
502 
503 	ret = user_path_at(path_fd, pathname, LOOKUP_FOLLOW, &path);
504 	if (ret)
505 		return ERR_PTR(ret);
506 
507 	inode = d_backing_inode(path.dentry);
508 	ret = path_permission(&path, ACC_MODE(flags));
509 	if (ret)
510 		goto out;
511 
512 	ret = bpf_inode_type(inode, type);
513 	if (ret)
514 		goto out;
515 
516 	raw = bpf_any_get(inode->i_private, *type);
517 	if (!IS_ERR(raw))
518 		touch_atime(&path);
519 
520 	path_put(&path);
521 	return raw;
522 out:
523 	path_put(&path);
524 	return ERR_PTR(ret);
525 }
526 
527 int bpf_obj_get_user(int path_fd, const char __user *pathname, int flags)
528 {
529 	enum bpf_type type = BPF_TYPE_UNSPEC;
530 	int f_flags;
531 	void *raw;
532 	int ret;
533 
534 	f_flags = bpf_get_file_flag(flags);
535 	if (f_flags < 0)
536 		return f_flags;
537 
538 	raw = bpf_obj_do_get(path_fd, pathname, &type, f_flags);
539 	if (IS_ERR(raw))
540 		return PTR_ERR(raw);
541 
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);
548 	else
549 		return -ENOENT;
550 
551 	if (ret < 0)
552 		bpf_any_put(raw, type);
553 	return ret;
554 }
555 
556 static struct bpf_prog *__get_prog_inode(struct inode *inode, enum bpf_prog_type type)
557 {
558 	struct bpf_prog *prog;
559 	int ret = inode_permission(&nop_mnt_idmap, inode, MAY_READ);
560 	if (ret)
561 		return ERR_PTR(ret);
562 
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);
569 
570 	prog = inode->i_private;
571 
572 	ret = security_bpf_prog(prog);
573 	if (ret < 0)
574 		return ERR_PTR(ret);
575 
576 	if (!bpf_prog_get_ok(prog, &type, false))
577 		return ERR_PTR(-EINVAL);
578 
579 	bpf_prog_inc(prog);
580 	return prog;
581 }
582 
583 struct bpf_prog *bpf_prog_get_type_path(const char *name, enum bpf_prog_type type)
584 {
585 	struct bpf_prog *prog;
586 	struct path path;
587 	int ret = kern_path(name, LOOKUP_FOLLOW, &path);
588 	if (ret)
589 		return ERR_PTR(ret);
590 	prog = __get_prog_inode(d_backing_inode(path.dentry), type);
591 	if (!IS_ERR(prog))
592 		touch_atime(&path);
593 	path_put(&path);
594 	return prog;
595 }
596 EXPORT_SYMBOL(bpf_prog_get_type_path);
597 
598 /*
599  * Display the mount options in /proc/mounts.
600  */
601 static int bpf_show_options(struct seq_file *m, struct dentry *root)
602 {
603 	struct bpf_mount_opts *opts = root->d_sb->s_fs_info;
604 	umode_t mode = d_inode(root)->i_mode & S_IALLUGO & ~S_ISVTX;
605 	u64 mask;
606 
607 	if (mode != S_IRWXUGO)
608 		seq_printf(m, ",mode=%o", mode);
609 
610 	mask = (1ULL << __MAX_BPF_CMD) - 1;
611 	if ((opts->delegate_cmds & mask) == mask)
612 		seq_printf(m, ",delegate_cmds=any");
613 	else if (opts->delegate_cmds)
614 		seq_printf(m, ",delegate_cmds=0x%llx", opts->delegate_cmds);
615 
616 	mask = (1ULL << __MAX_BPF_MAP_TYPE) - 1;
617 	if ((opts->delegate_maps & mask) == mask)
618 		seq_printf(m, ",delegate_maps=any");
619 	else if (opts->delegate_maps)
620 		seq_printf(m, ",delegate_maps=0x%llx", opts->delegate_maps);
621 
622 	mask = (1ULL << __MAX_BPF_PROG_TYPE) - 1;
623 	if ((opts->delegate_progs & mask) == mask)
624 		seq_printf(m, ",delegate_progs=any");
625 	else if (opts->delegate_progs)
626 		seq_printf(m, ",delegate_progs=0x%llx", opts->delegate_progs);
627 
628 	mask = (1ULL << __MAX_BPF_ATTACH_TYPE) - 1;
629 	if ((opts->delegate_attachs & mask) == mask)
630 		seq_printf(m, ",delegate_attachs=any");
631 	else if (opts->delegate_attachs)
632 		seq_printf(m, ",delegate_attachs=0x%llx", opts->delegate_attachs);
633 	return 0;
634 }
635 
636 static void bpf_free_inode(struct inode *inode)
637 {
638 	enum bpf_type type;
639 
640 	if (S_ISLNK(inode->i_mode))
641 		kfree(inode->i_link);
642 	if (!bpf_inode_type(inode, &type))
643 		bpf_any_put(inode->i_private, type);
644 	free_inode_nonrcu(inode);
645 }
646 
647 const struct super_operations bpf_super_ops = {
648 	.statfs		= simple_statfs,
649 	.drop_inode	= generic_delete_inode,
650 	.show_options	= bpf_show_options,
651 	.free_inode	= bpf_free_inode,
652 };
653 
654 enum {
655 	OPT_MODE,
656 	OPT_DELEGATE_CMDS,
657 	OPT_DELEGATE_MAPS,
658 	OPT_DELEGATE_PROGS,
659 	OPT_DELEGATE_ATTACHS,
660 };
661 
662 static const struct fs_parameter_spec bpf_fs_parameters[] = {
663 	fsparam_u32oct	("mode",			OPT_MODE),
664 	fsparam_string	("delegate_cmds",		OPT_DELEGATE_CMDS),
665 	fsparam_string	("delegate_maps",		OPT_DELEGATE_MAPS),
666 	fsparam_string	("delegate_progs",		OPT_DELEGATE_PROGS),
667 	fsparam_string	("delegate_attachs",		OPT_DELEGATE_ATTACHS),
668 	{}
669 };
670 
671 static int bpf_parse_param(struct fs_context *fc, struct fs_parameter *param)
672 {
673 	struct bpf_mount_opts *opts = fc->s_fs_info;
674 	struct fs_parse_result result;
675 	int opt, err;
676 	u64 msk;
677 
678 	opt = fs_parse(fc, bpf_fs_parameters, param, &result);
679 	if (opt < 0) {
680 		/* We might like to report bad mount options here, but
681 		 * traditionally we've ignored all mount options, so we'd
682 		 * better continue to ignore non-existing options for bpf.
683 		 */
684 		if (opt == -ENOPARAM) {
685 			opt = vfs_parse_fs_param_source(fc, param);
686 			if (opt != -ENOPARAM)
687 				return opt;
688 
689 			return 0;
690 		}
691 
692 		if (opt < 0)
693 			return opt;
694 	}
695 
696 	switch (opt) {
697 	case OPT_MODE:
698 		opts->mode = result.uint_32 & S_IALLUGO;
699 		break;
700 	case OPT_DELEGATE_CMDS:
701 	case OPT_DELEGATE_MAPS:
702 	case OPT_DELEGATE_PROGS:
703 	case OPT_DELEGATE_ATTACHS:
704 		if (strcmp(param->string, "any") == 0) {
705 			msk = ~0ULL;
706 		} else {
707 			err = kstrtou64(param->string, 0, &msk);
708 			if (err)
709 				return err;
710 		}
711 		/* Setting delegation mount options requires privileges */
712 		if (msk && !capable(CAP_SYS_ADMIN))
713 			return -EPERM;
714 		switch (opt) {
715 		case OPT_DELEGATE_CMDS: opts->delegate_cmds |= msk; break;
716 		case OPT_DELEGATE_MAPS: opts->delegate_maps |= msk; break;
717 		case OPT_DELEGATE_PROGS: opts->delegate_progs |= msk; break;
718 		case OPT_DELEGATE_ATTACHS: opts->delegate_attachs |= msk; break;
719 		default: return -EINVAL;
720 		}
721 		break;
722 	}
723 
724 	return 0;
725 }
726 
727 struct bpf_preload_ops *bpf_preload_ops;
728 EXPORT_SYMBOL_GPL(bpf_preload_ops);
729 
730 static bool bpf_preload_mod_get(void)
731 {
732 	/* If bpf_preload.ko wasn't loaded earlier then load it now.
733 	 * When bpf_preload is built into vmlinux the module's __init
734 	 * function will populate it.
735 	 */
736 	if (!bpf_preload_ops) {
737 		request_module("bpf_preload");
738 		if (!bpf_preload_ops)
739 			return false;
740 	}
741 	/* And grab the reference, so the module doesn't disappear while the
742 	 * kernel is interacting with the kernel module and its UMD.
743 	 */
744 	if (!try_module_get(bpf_preload_ops->owner)) {
745 		pr_err("bpf_preload module get failed.\n");
746 		return false;
747 	}
748 	return true;
749 }
750 
751 static void bpf_preload_mod_put(void)
752 {
753 	if (bpf_preload_ops)
754 		/* now user can "rmmod bpf_preload" if necessary */
755 		module_put(bpf_preload_ops->owner);
756 }
757 
758 static DEFINE_MUTEX(bpf_preload_lock);
759 
760 static int populate_bpffs(struct dentry *parent)
761 {
762 	struct bpf_preload_info objs[BPF_PRELOAD_LINKS] = {};
763 	int err = 0, i;
764 
765 	/* grab the mutex to make sure the kernel interactions with bpf_preload
766 	 * are serialized
767 	 */
768 	mutex_lock(&bpf_preload_lock);
769 
770 	/* if bpf_preload.ko wasn't built into vmlinux then load it */
771 	if (!bpf_preload_mod_get())
772 		goto out;
773 
774 	err = bpf_preload_ops->preload(objs);
775 	if (err)
776 		goto out_put;
777 	for (i = 0; i < BPF_PRELOAD_LINKS; i++) {
778 		bpf_link_inc(objs[i].link);
779 		err = bpf_iter_link_pin_kernel(parent,
780 					       objs[i].link_name, objs[i].link);
781 		if (err) {
782 			bpf_link_put(objs[i].link);
783 			goto out_put;
784 		}
785 	}
786 out_put:
787 	bpf_preload_mod_put();
788 out:
789 	mutex_unlock(&bpf_preload_lock);
790 	return err;
791 }
792 
793 static int bpf_fill_super(struct super_block *sb, struct fs_context *fc)
794 {
795 	static const struct tree_descr bpf_rfiles[] = { { "" } };
796 	struct bpf_mount_opts *opts = sb->s_fs_info;
797 	struct inode *inode;
798 	int ret;
799 
800 	/* Mounting an instance of BPF FS requires privileges */
801 	if (fc->user_ns != &init_user_ns && !capable(CAP_SYS_ADMIN))
802 		return -EPERM;
803 
804 	ret = simple_fill_super(sb, BPF_FS_MAGIC, bpf_rfiles);
805 	if (ret)
806 		return ret;
807 
808 	sb->s_op = &bpf_super_ops;
809 
810 	inode = sb->s_root->d_inode;
811 	inode->i_op = &bpf_dir_iops;
812 	inode->i_mode &= ~S_IALLUGO;
813 	populate_bpffs(sb->s_root);
814 	inode->i_mode |= S_ISVTX | opts->mode;
815 	return 0;
816 }
817 
818 static int bpf_get_tree(struct fs_context *fc)
819 {
820 	return get_tree_nodev(fc, bpf_fill_super);
821 }
822 
823 static void bpf_free_fc(struct fs_context *fc)
824 {
825 	kfree(fc->s_fs_info);
826 }
827 
828 static const struct fs_context_operations bpf_context_ops = {
829 	.free		= bpf_free_fc,
830 	.parse_param	= bpf_parse_param,
831 	.get_tree	= bpf_get_tree,
832 };
833 
834 /*
835  * Set up the filesystem mount context.
836  */
837 static int bpf_init_fs_context(struct fs_context *fc)
838 {
839 	struct bpf_mount_opts *opts;
840 
841 	opts = kzalloc(sizeof(struct bpf_mount_opts), GFP_KERNEL);
842 	if (!opts)
843 		return -ENOMEM;
844 
845 	opts->mode = S_IRWXUGO;
846 
847 	/* start out with no BPF token delegation enabled */
848 	opts->delegate_cmds = 0;
849 	opts->delegate_maps = 0;
850 	opts->delegate_progs = 0;
851 	opts->delegate_attachs = 0;
852 
853 	fc->s_fs_info = opts;
854 	fc->ops = &bpf_context_ops;
855 	return 0;
856 }
857 
858 static void bpf_kill_super(struct super_block *sb)
859 {
860 	struct bpf_mount_opts *opts = sb->s_fs_info;
861 
862 	kill_litter_super(sb);
863 	kfree(opts);
864 }
865 
866 static struct file_system_type bpf_fs_type = {
867 	.owner		= THIS_MODULE,
868 	.name		= "bpf",
869 	.init_fs_context = bpf_init_fs_context,
870 	.parameters	= bpf_fs_parameters,
871 	.kill_sb	= bpf_kill_super,
872 	.fs_flags	= FS_USERNS_MOUNT,
873 };
874 
875 static int __init bpf_init(void)
876 {
877 	int ret;
878 
879 	ret = sysfs_create_mount_point(fs_kobj, "bpf");
880 	if (ret)
881 		return ret;
882 
883 	ret = register_filesystem(&bpf_fs_type);
884 	if (ret)
885 		sysfs_remove_mount_point(fs_kobj, "bpf");
886 
887 	return ret;
888 }
889 fs_initcall(bpf_init);
890