Linux Audio

Check our new training course

Loading...
v6.2
  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
 25enum bpf_type {
 26	BPF_TYPE_UNSPEC	= 0,
 27	BPF_TYPE_PROG,
 28	BPF_TYPE_MAP,
 29	BPF_TYPE_LINK,
 30};
 31
 32static 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
 52static 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
 70static 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
 95static const struct inode_operations bpf_dir_iops;
 96
 97static const struct inode_operations bpf_prog_iops = { };
 98static const struct inode_operations bpf_map_iops  = { };
 99static const struct inode_operations bpf_link_iops  = { };
100
101static 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	inode->i_atime = current_time(inode);
122	inode->i_mtime = inode->i_atime;
123	inode->i_ctime = inode->i_atime;
124
125	inode_init_owner(&init_user_ns, inode, dir, mode);
126
127	return inode;
128}
129
130static int bpf_inode_type(const struct inode *inode, enum bpf_type *type)
131{
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;
141
142	return 0;
143}
144
145static void bpf_dentry_finalize(struct dentry *dentry, struct inode *inode,
146				struct inode *dir)
147{
148	d_instantiate(dentry, inode);
149	dget(dentry);
150
151	dir->i_mtime = current_time(dir);
152	dir->i_ctime = dir->i_mtime;
153}
154
155static int bpf_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
156		     struct dentry *dentry, umode_t mode)
157{
158	struct inode *inode;
159
160	inode = bpf_get_inode(dir->i_sb, dir, mode | S_IFDIR);
161	if (IS_ERR(inode))
162		return PTR_ERR(inode);
163
164	inode->i_op = &bpf_dir_iops;
165	inode->i_fop = &simple_dir_operations;
166
167	inc_nlink(inode);
168	inc_nlink(dir);
169
170	bpf_dentry_finalize(dentry, inode, dir);
171	return 0;
172}
173
174struct map_iter {
175	void *key;
176	bool done;
177};
178
179static struct map_iter *map_iter(struct seq_file *m)
180{
181	return m->private;
182}
183
184static struct bpf_map *seq_file_to_map(struct seq_file *m)
185{
186	return file_inode(m->file)->i_private;
187}
188
189static void map_iter_free(struct map_iter *iter)
190{
191	if (iter) {
192		kfree(iter->key);
193		kfree(iter);
194	}
195}
196
197static struct map_iter *map_iter_alloc(struct bpf_map *map)
198{
199	struct map_iter *iter;
200
201	iter = kzalloc(sizeof(*iter), GFP_KERNEL | __GFP_NOWARN);
202	if (!iter)
203		goto error;
204
205	iter->key = kzalloc(map->key_size, GFP_KERNEL | __GFP_NOWARN);
206	if (!iter->key)
207		goto error;
208
209	return iter;
210
211error:
212	map_iter_free(iter);
213	return NULL;
214}
215
216static void *map_seq_next(struct seq_file *m, void *v, loff_t *pos)
217{
218	struct bpf_map *map = seq_file_to_map(m);
219	void *key = map_iter(m)->key;
220	void *prev_key;
221
222	(*pos)++;
223	if (map_iter(m)->done)
224		return NULL;
225
226	if (unlikely(v == SEQ_START_TOKEN))
227		prev_key = NULL;
228	else
229		prev_key = key;
230
231	rcu_read_lock();
232	if (map->ops->map_get_next_key(map, prev_key, key)) {
233		map_iter(m)->done = true;
234		key = NULL;
235	}
236	rcu_read_unlock();
237	return key;
238}
239
240static void *map_seq_start(struct seq_file *m, loff_t *pos)
241{
242	if (map_iter(m)->done)
243		return NULL;
244
245	return *pos ? map_iter(m)->key : SEQ_START_TOKEN;
246}
247
248static void map_seq_stop(struct seq_file *m, void *v)
249{
250}
251
252static int map_seq_show(struct seq_file *m, void *v)
253{
254	struct bpf_map *map = seq_file_to_map(m);
255	void *key = map_iter(m)->key;
256
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");
260	} else {
261		map->ops->map_seq_show_elem(map, key, m);
262	}
263
264	return 0;
265}
266
267static 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,
272};
273
274static int bpffs_map_open(struct inode *inode, struct file *file)
275{
276	struct bpf_map *map = inode->i_private;
277	struct map_iter *iter;
278	struct seq_file *m;
279	int err;
280
281	iter = map_iter_alloc(map);
282	if (!iter)
283		return -ENOMEM;
284
285	err = seq_open(file, &bpffs_map_seq_ops);
286	if (err) {
287		map_iter_free(iter);
288		return err;
289	}
290
291	m = file->private_data;
292	m->private = iter;
293
294	return 0;
295}
296
297static int bpffs_map_release(struct inode *inode, struct file *file)
298{
299	struct seq_file *m = file->private_data;
300
301	map_iter_free(map_iter(m));
302
303	return seq_release(inode, file);
304}
305
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".
310 *
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
314 * interface.
315 */
316static const struct file_operations bpffs_map_fops = {
317	.open		= bpffs_map_open,
318	.read		= seq_read,
319	.release	= bpffs_map_release,
320};
321
322static int bpffs_obj_open(struct inode *inode, struct file *file)
323{
324	return -EIO;
325}
326
327static const struct file_operations bpffs_obj_fops = {
328	.open		= bpffs_obj_open,
329};
330
331static int bpf_mkobj_ops(struct dentry *dentry, umode_t mode, void *raw,
332			 const struct inode_operations *iops,
333			 const struct file_operations *fops)
334{
335	struct inode *dir = dentry->d_parent->d_inode;
336	struct inode *inode = bpf_get_inode(dir->i_sb, dir, mode);
337	if (IS_ERR(inode))
338		return PTR_ERR(inode);
339
340	inode->i_op = iops;
341	inode->i_fop = fops;
342	inode->i_private = raw;
343
344	bpf_dentry_finalize(dentry, inode, dir);
345	return 0;
346}
347
348static int bpf_mkprog(struct dentry *dentry, umode_t mode, void *arg)
349{
350	return bpf_mkobj_ops(dentry, mode, arg, &bpf_prog_iops,
351			     &bpffs_obj_fops);
352}
353
354static int bpf_mkmap(struct dentry *dentry, umode_t mode, void *arg)
355{
356	struct bpf_map *map = arg;
357
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);
361}
362
363static int bpf_mklink(struct dentry *dentry, umode_t mode, void *arg)
364{
365	struct bpf_link *link = arg;
366
367	return bpf_mkobj_ops(dentry, mode, arg, &bpf_link_iops,
368			     bpf_link_is_iter(link) ?
369			     &bpf_iter_fops : &bpffs_obj_fops);
370}
371
372static struct dentry *
373bpf_lookup(struct inode *dir, struct dentry *dentry, unsigned flags)
374{
375	/* Dots in names (e.g. "/sys/fs/bpf/foo.bar") are reserved for future
376	 * extensions. That allows popoulate_bpffs() create special files.
377	 */
378	if ((dir->i_mode & S_IALLUGO) &&
379	    strchr(dentry->d_name.name, '.'))
380		return ERR_PTR(-EPERM);
381
382	return simple_lookup(dir, dentry, flags);
383}
384
385static int bpf_symlink(struct user_namespace *mnt_userns, struct inode *dir,
386		       struct dentry *dentry, const char *target)
387{
388	char *link = kstrdup(target, GFP_USER | __GFP_NOWARN);
389	struct inode *inode;
390
391	if (!link)
392		return -ENOMEM;
393
394	inode = bpf_get_inode(dir->i_sb, dir, S_IRWXUGO | S_IFLNK);
395	if (IS_ERR(inode)) {
396		kfree(link);
397		return PTR_ERR(inode);
398	}
399
400	inode->i_op = &simple_symlink_inode_operations;
401	inode->i_link = link;
402
403	bpf_dentry_finalize(dentry, inode, dir);
404	return 0;
405}
406
407static const struct inode_operations bpf_dir_iops = {
408	.lookup		= bpf_lookup,
409	.mkdir		= bpf_mkdir,
410	.symlink	= bpf_symlink,
411	.rmdir		= simple_rmdir,
412	.rename		= simple_rename,
413	.link		= simple_link,
414	.unlink		= simple_unlink,
415};
416
417/* pin iterator link into bpffs */
418static int bpf_iter_link_pin_kernel(struct dentry *parent,
419				    const char *name, struct bpf_link *link)
420{
421	umode_t mode = S_IFREG | S_IRUSR;
422	struct dentry *dentry;
423	int ret;
424
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);
430	}
431	ret = bpf_mkobj_ops(dentry, mode, link, &bpf_link_iops,
432			    &bpf_iter_fops);
433	dput(dentry);
434	inode_unlock(parent->d_inode);
435	return ret;
436}
437
438static int bpf_obj_do_pin(const char __user *pathname, void *raw,
439			  enum bpf_type type)
440{
441	struct dentry *dentry;
442	struct inode *dir;
443	struct path path;
444	umode_t mode;
445	int ret;
446
447	dentry = user_path_create(AT_FDCWD, pathname, &path, 0);
448	if (IS_ERR(dentry))
449		return PTR_ERR(dentry);
450
451	mode = S_IFREG | ((S_IRUSR | S_IWUSR) & ~current_umask());
452
453	ret = security_path_mknod(&path, dentry, mode, 0);
454	if (ret)
455		goto out;
456
457	dir = d_inode(path.dentry);
458	if (dir->i_op != &bpf_dir_iops) {
459		ret = -EPERM;
460		goto out;
461	}
462
 
 
 
 
 
463	switch (type) {
464	case BPF_TYPE_PROG:
465		ret = vfs_mkobj(dentry, mode, bpf_mkprog, raw);
466		break;
467	case BPF_TYPE_MAP:
468		ret = vfs_mkobj(dentry, mode, bpf_mkmap, raw);
469		break;
470	case BPF_TYPE_LINK:
471		ret = vfs_mkobj(dentry, mode, bpf_mklink, raw);
472		break;
473	default:
474		ret = -EPERM;
475	}
476out:
477	done_path_create(&path, dentry);
478	return ret;
479}
480
481int bpf_obj_pin_user(u32 ufd, const char __user *pathname)
482{
483	enum bpf_type type;
484	void *raw;
485	int ret;
486
487	raw = bpf_fd_probe_obj(ufd, &type);
488	if (IS_ERR(raw))
489		return PTR_ERR(raw);
490
491	ret = bpf_obj_do_pin(pathname, raw, type);
492	if (ret != 0)
493		bpf_any_put(raw, type);
494
495	return ret;
496}
497
498static void *bpf_obj_do_get(const char __user *pathname,
499			    enum bpf_type *type, int flags)
500{
501	struct inode *inode;
502	struct path path;
503	void *raw;
504	int ret;
505
506	ret = user_path_at(AT_FDCWD, pathname, LOOKUP_FOLLOW, &path);
507	if (ret)
508		return ERR_PTR(ret);
509
510	inode = d_backing_inode(path.dentry);
511	ret = path_permission(&path, ACC_MODE(flags));
512	if (ret)
513		goto out;
514
515	ret = bpf_inode_type(inode, type);
516	if (ret)
517		goto out;
518
519	raw = bpf_any_get(inode->i_private, *type);
520	if (!IS_ERR(raw))
521		touch_atime(&path);
522
523	path_put(&path);
524	return raw;
525out:
526	path_put(&path);
527	return ERR_PTR(ret);
528}
529
530int bpf_obj_get_user(const char __user *pathname, int flags)
531{
532	enum bpf_type type = BPF_TYPE_UNSPEC;
533	int f_flags;
534	void *raw;
535	int ret;
536
537	f_flags = bpf_get_file_flag(flags);
538	if (f_flags < 0)
539		return f_flags;
540
541	raw = bpf_obj_do_get(pathname, &type, f_flags);
542	if (IS_ERR(raw))
543		return PTR_ERR(raw);
544
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);
551	else
552		return -ENOENT;
553
554	if (ret < 0)
555		bpf_any_put(raw, type);
556	return ret;
557}
558
559static struct bpf_prog *__get_prog_inode(struct inode *inode, enum bpf_prog_type type)
560{
561	struct bpf_prog *prog;
562	int ret = inode_permission(&init_user_ns, inode, MAY_READ);
563	if (ret)
564		return ERR_PTR(ret);
565
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);
572
573	prog = inode->i_private;
574
575	ret = security_bpf_prog(prog);
576	if (ret < 0)
577		return ERR_PTR(ret);
578
579	if (!bpf_prog_get_ok(prog, &type, false))
580		return ERR_PTR(-EINVAL);
581
582	bpf_prog_inc(prog);
583	return prog;
584}
585
586struct bpf_prog *bpf_prog_get_type_path(const char *name, enum bpf_prog_type type)
587{
588	struct bpf_prog *prog;
589	struct path path;
590	int ret = kern_path(name, LOOKUP_FOLLOW, &path);
591	if (ret)
592		return ERR_PTR(ret);
593	prog = __get_prog_inode(d_backing_inode(path.dentry), type);
594	if (!IS_ERR(prog))
595		touch_atime(&path);
596	path_put(&path);
597	return prog;
598}
599EXPORT_SYMBOL(bpf_prog_get_type_path);
600
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
601/*
602 * Display the mount options in /proc/mounts.
603 */
604static int bpf_show_options(struct seq_file *m, struct dentry *root)
605{
606	umode_t mode = d_inode(root)->i_mode & S_IALLUGO & ~S_ISVTX;
607
 
 
 
 
 
 
 
 
 
608	if (mode != S_IRWXUGO)
609		seq_printf(m, ",mode=%o", mode);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
610	return 0;
611}
612
613static void bpf_free_inode(struct inode *inode)
614{
615	enum bpf_type type;
616
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);
622}
623
624static 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,
629};
630
631enum {
 
 
632	OPT_MODE,
 
 
 
 
633};
634
635static const struct fs_parameter_spec bpf_fs_parameters[] = {
 
 
636	fsparam_u32oct	("mode",			OPT_MODE),
 
 
 
 
637	{}
638};
639
640struct bpf_mount_opts {
641	umode_t mode;
642};
643
644static int bpf_parse_param(struct fs_context *fc, struct fs_parameter *param)
645{
646	struct bpf_mount_opts *opts = fc->fs_private;
647	struct fs_parse_result result;
648	int opt;
 
 
649
650	opt = fs_parse(fc, bpf_fs_parameters, param, &result);
651	if (opt < 0) {
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.
655		 */
656		if (opt == -ENOPARAM) {
657			opt = vfs_parse_fs_param_source(fc, param);
658			if (opt != -ENOPARAM)
659				return opt;
660
661			return 0;
662		}
663
664		if (opt < 0)
665			return opt;
666	}
667
668	switch (opt) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
669	case OPT_MODE:
670		opts->mode = result.uint_32 & S_IALLUGO;
671		break;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
672	}
673
674	return 0;
 
 
675}
676
677struct bpf_preload_ops *bpf_preload_ops;
678EXPORT_SYMBOL_GPL(bpf_preload_ops);
679
680static bool bpf_preload_mod_get(void)
681{
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.
685	 */
686	if (!bpf_preload_ops) {
687		request_module("bpf_preload");
688		if (!bpf_preload_ops)
689			return false;
690	}
691	/* And grab the reference, so the module doesn't disappear while the
692	 * kernel is interacting with the kernel module and its UMD.
693	 */
694	if (!try_module_get(bpf_preload_ops->owner)) {
695		pr_err("bpf_preload module get failed.\n");
696		return false;
697	}
698	return true;
699}
700
701static void bpf_preload_mod_put(void)
702{
703	if (bpf_preload_ops)
704		/* now user can "rmmod bpf_preload" if necessary */
705		module_put(bpf_preload_ops->owner);
706}
707
708static DEFINE_MUTEX(bpf_preload_lock);
709
710static int populate_bpffs(struct dentry *parent)
711{
712	struct bpf_preload_info objs[BPF_PRELOAD_LINKS] = {};
713	int err = 0, i;
714
715	/* grab the mutex to make sure the kernel interactions with bpf_preload
716	 * are serialized
717	 */
718	mutex_lock(&bpf_preload_lock);
719
720	/* if bpf_preload.ko wasn't built into vmlinux then load it */
721	if (!bpf_preload_mod_get())
722		goto out;
723
724	err = bpf_preload_ops->preload(objs);
725	if (err)
726		goto out_put;
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);
731		if (err) {
732			bpf_link_put(objs[i].link);
733			goto out_put;
734		}
735	}
736out_put:
737	bpf_preload_mod_put();
738out:
739	mutex_unlock(&bpf_preload_lock);
740	return err;
741}
742
743static int bpf_fill_super(struct super_block *sb, struct fs_context *fc)
744{
745	static const struct tree_descr bpf_rfiles[] = { { "" } };
746	struct bpf_mount_opts *opts = fc->fs_private;
747	struct inode *inode;
748	int ret;
749
 
 
 
 
750	ret = simple_fill_super(sb, BPF_FS_MAGIC, bpf_rfiles);
751	if (ret)
752		return ret;
753
754	sb->s_op = &bpf_super_ops;
755
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;
761	return 0;
762}
763
764static int bpf_get_tree(struct fs_context *fc)
765{
766	return get_tree_nodev(fc, bpf_fill_super);
767}
768
769static void bpf_free_fc(struct fs_context *fc)
770{
771	kfree(fc->fs_private);
772}
773
774static const struct fs_context_operations bpf_context_ops = {
775	.free		= bpf_free_fc,
776	.parse_param	= bpf_parse_param,
777	.get_tree	= bpf_get_tree,
778};
779
780/*
781 * Set up the filesystem mount context.
782 */
783static int bpf_init_fs_context(struct fs_context *fc)
784{
785	struct bpf_mount_opts *opts;
786
787	opts = kzalloc(sizeof(struct bpf_mount_opts), GFP_KERNEL);
788	if (!opts)
789		return -ENOMEM;
790
791	opts->mode = S_IRWXUGO;
 
 
 
 
 
 
 
 
792
793	fc->fs_private = opts;
794	fc->ops = &bpf_context_ops;
795	return 0;
796}
797
 
 
 
 
 
 
 
 
798static struct file_system_type bpf_fs_type = {
799	.owner		= THIS_MODULE,
800	.name		= "bpf",
801	.init_fs_context = bpf_init_fs_context,
802	.parameters	= bpf_fs_parameters,
803	.kill_sb	= kill_litter_super,
 
804};
805
806static int __init bpf_init(void)
807{
808	int ret;
809
810	ret = sysfs_create_mount_point(fs_kobj, "bpf");
811	if (ret)
812		return ret;
813
814	ret = register_filesystem(&bpf_fs_type);
815	if (ret)
816		sysfs_remove_mount_point(fs_kobj, "bpf");
817
818	return ret;
819}
820fs_initcall(bpf_init);
v6.9.4
   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 <linux/kstrtox.h>
  24#include "preload/bpf_preload.h"
  25
  26enum bpf_type {
  27	BPF_TYPE_UNSPEC	= 0,
  28	BPF_TYPE_PROG,
  29	BPF_TYPE_MAP,
  30	BPF_TYPE_LINK,
  31};
  32
  33static 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
  53static 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
  71static 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
  96static const struct inode_operations bpf_dir_iops;
  97
  98static const struct inode_operations bpf_prog_iops = { };
  99static const struct inode_operations bpf_map_iops  = { };
 100static const struct inode_operations bpf_link_iops  = { };
 101
 102struct 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
 129static 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
 144static 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
 153static 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
 172struct map_iter {
 173	void *key;
 174	bool done;
 175};
 176
 177static struct map_iter *map_iter(struct seq_file *m)
 178{
 179	return m->private;
 180}
 181
 182static struct bpf_map *seq_file_to_map(struct seq_file *m)
 183{
 184	return file_inode(m->file)->i_private;
 185}
 186
 187static void map_iter_free(struct map_iter *iter)
 188{
 189	if (iter) {
 190		kfree(iter->key);
 191		kfree(iter);
 192	}
 193}
 194
 195static 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
 209error:
 210	map_iter_free(iter);
 211	return NULL;
 212}
 213
 214static 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
 238static 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
 246static void map_seq_stop(struct seq_file *m, void *v)
 247{
 248}
 249
 250static 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
 265static 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
 272static 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
 295static 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 */
 314static const struct file_operations bpffs_map_fops = {
 315	.open		= bpffs_map_open,
 316	.read		= seq_read,
 317	.release	= bpffs_map_release,
 318};
 319
 320static int bpffs_obj_open(struct inode *inode, struct file *file)
 321{
 322	return -EIO;
 323}
 324
 325static const struct file_operations bpffs_obj_fops = {
 326	.open		= bpffs_obj_open,
 327};
 328
 329static 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
 346static 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
 352static 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
 361static 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
 370static struct dentry *
 371bpf_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
 383static 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
 405static 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 */
 416static 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
 436static 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	}
 473out:
 474	done_path_create(&path, dentry);
 475	return ret;
 476}
 477
 478int 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
 495static 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;
 522out:
 523	path_put(&path);
 524	return ERR_PTR(ret);
 525}
 526
 527int 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
 556static 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
 583struct 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}
 596EXPORT_SYMBOL(bpf_prog_get_type_path);
 597
 598struct bpffs_btf_enums {
 599	const struct btf *btf;
 600	const struct btf_type *cmd_t;
 601	const struct btf_type *map_t;
 602	const struct btf_type *prog_t;
 603	const struct btf_type *attach_t;
 604};
 605
 606static int find_bpffs_btf_enums(struct bpffs_btf_enums *info)
 607{
 608	const struct btf *btf;
 609	const struct btf_type *t;
 610	const char *name;
 611	int i, n;
 612
 613	memset(info, 0, sizeof(*info));
 614
 615	btf = bpf_get_btf_vmlinux();
 616	if (IS_ERR(btf))
 617		return PTR_ERR(btf);
 618	if (!btf)
 619		return -ENOENT;
 620
 621	info->btf = btf;
 622
 623	for (i = 1, n = btf_nr_types(btf); i < n; i++) {
 624		t = btf_type_by_id(btf, i);
 625		if (!btf_type_is_enum(t))
 626			continue;
 627
 628		name = btf_name_by_offset(btf, t->name_off);
 629		if (!name)
 630			continue;
 631
 632		if (strcmp(name, "bpf_cmd") == 0)
 633			info->cmd_t = t;
 634		else if (strcmp(name, "bpf_map_type") == 0)
 635			info->map_t = t;
 636		else if (strcmp(name, "bpf_prog_type") == 0)
 637			info->prog_t = t;
 638		else if (strcmp(name, "bpf_attach_type") == 0)
 639			info->attach_t = t;
 640		else
 641			continue;
 642
 643		if (info->cmd_t && info->map_t && info->prog_t && info->attach_t)
 644			return 0;
 645	}
 646
 647	return -ESRCH;
 648}
 649
 650static bool find_btf_enum_const(const struct btf *btf, const struct btf_type *enum_t,
 651				const char *prefix, const char *str, int *value)
 652{
 653	const struct btf_enum *e;
 654	const char *name;
 655	int i, n, pfx_len = strlen(prefix);
 656
 657	*value = 0;
 658
 659	if (!btf || !enum_t)
 660		return false;
 661
 662	for (i = 0, n = btf_vlen(enum_t); i < n; i++) {
 663		e = &btf_enum(enum_t)[i];
 664
 665		name = btf_name_by_offset(btf, e->name_off);
 666		if (!name || strncasecmp(name, prefix, pfx_len) != 0)
 667			continue;
 668
 669		/* match symbolic name case insensitive and ignoring prefix */
 670		if (strcasecmp(name + pfx_len, str) == 0) {
 671			*value = e->val;
 672			return true;
 673		}
 674	}
 675
 676	return false;
 677}
 678
 679static void seq_print_delegate_opts(struct seq_file *m,
 680				    const char *opt_name,
 681				    const struct btf *btf,
 682				    const struct btf_type *enum_t,
 683				    const char *prefix,
 684				    u64 delegate_msk, u64 any_msk)
 685{
 686	const struct btf_enum *e;
 687	bool first = true;
 688	const char *name;
 689	u64 msk;
 690	int i, n, pfx_len = strlen(prefix);
 691
 692	delegate_msk &= any_msk; /* clear unknown bits */
 693
 694	if (delegate_msk == 0)
 695		return;
 696
 697	seq_printf(m, ",%s", opt_name);
 698	if (delegate_msk == any_msk) {
 699		seq_printf(m, "=any");
 700		return;
 701	}
 702
 703	if (btf && enum_t) {
 704		for (i = 0, n = btf_vlen(enum_t); i < n; i++) {
 705			e = &btf_enum(enum_t)[i];
 706			name = btf_name_by_offset(btf, e->name_off);
 707			if (!name || strncasecmp(name, prefix, pfx_len) != 0)
 708				continue;
 709			msk = 1ULL << e->val;
 710			if (delegate_msk & msk) {
 711				/* emit lower-case name without prefix */
 712				seq_printf(m, "%c", first ? '=' : ':');
 713				name += pfx_len;
 714				while (*name) {
 715					seq_printf(m, "%c", tolower(*name));
 716					name++;
 717				}
 718
 719				delegate_msk &= ~msk;
 720				first = false;
 721			}
 722		}
 723	}
 724	if (delegate_msk)
 725		seq_printf(m, "%c0x%llx", first ? '=' : ':', delegate_msk);
 726}
 727
 728/*
 729 * Display the mount options in /proc/mounts.
 730 */
 731static int bpf_show_options(struct seq_file *m, struct dentry *root)
 732{
 733	struct inode *inode = d_inode(root);
 734	umode_t mode = inode->i_mode & S_IALLUGO & ~S_ISVTX;
 735	struct bpf_mount_opts *opts = root->d_sb->s_fs_info;
 736	u64 mask;
 737
 738	if (!uid_eq(inode->i_uid, GLOBAL_ROOT_UID))
 739		seq_printf(m, ",uid=%u",
 740			   from_kuid_munged(&init_user_ns, inode->i_uid));
 741	if (!gid_eq(inode->i_gid, GLOBAL_ROOT_GID))
 742		seq_printf(m, ",gid=%u",
 743			   from_kgid_munged(&init_user_ns, inode->i_gid));
 744	if (mode != S_IRWXUGO)
 745		seq_printf(m, ",mode=%o", mode);
 746
 747	if (opts->delegate_cmds || opts->delegate_maps ||
 748	    opts->delegate_progs || opts->delegate_attachs) {
 749		struct bpffs_btf_enums info;
 750
 751		/* ignore errors, fallback to hex */
 752		(void)find_bpffs_btf_enums(&info);
 753
 754		mask = (1ULL << __MAX_BPF_CMD) - 1;
 755		seq_print_delegate_opts(m, "delegate_cmds",
 756					info.btf, info.cmd_t, "BPF_",
 757					opts->delegate_cmds, mask);
 758
 759		mask = (1ULL << __MAX_BPF_MAP_TYPE) - 1;
 760		seq_print_delegate_opts(m, "delegate_maps",
 761					info.btf, info.map_t, "BPF_MAP_TYPE_",
 762					opts->delegate_maps, mask);
 763
 764		mask = (1ULL << __MAX_BPF_PROG_TYPE) - 1;
 765		seq_print_delegate_opts(m, "delegate_progs",
 766					info.btf, info.prog_t, "BPF_PROG_TYPE_",
 767					opts->delegate_progs, mask);
 768
 769		mask = (1ULL << __MAX_BPF_ATTACH_TYPE) - 1;
 770		seq_print_delegate_opts(m, "delegate_attachs",
 771					info.btf, info.attach_t, "BPF_",
 772					opts->delegate_attachs, mask);
 773	}
 774
 775	return 0;
 776}
 777
 778static void bpf_free_inode(struct inode *inode)
 779{
 780	enum bpf_type type;
 781
 782	if (S_ISLNK(inode->i_mode))
 783		kfree(inode->i_link);
 784	if (!bpf_inode_type(inode, &type))
 785		bpf_any_put(inode->i_private, type);
 786	free_inode_nonrcu(inode);
 787}
 788
 789const struct super_operations bpf_super_ops = {
 790	.statfs		= simple_statfs,
 791	.drop_inode	= generic_delete_inode,
 792	.show_options	= bpf_show_options,
 793	.free_inode	= bpf_free_inode,
 794};
 795
 796enum {
 797	OPT_UID,
 798	OPT_GID,
 799	OPT_MODE,
 800	OPT_DELEGATE_CMDS,
 801	OPT_DELEGATE_MAPS,
 802	OPT_DELEGATE_PROGS,
 803	OPT_DELEGATE_ATTACHS,
 804};
 805
 806static const struct fs_parameter_spec bpf_fs_parameters[] = {
 807	fsparam_u32	("uid",				OPT_UID),
 808	fsparam_u32	("gid",				OPT_GID),
 809	fsparam_u32oct	("mode",			OPT_MODE),
 810	fsparam_string	("delegate_cmds",		OPT_DELEGATE_CMDS),
 811	fsparam_string	("delegate_maps",		OPT_DELEGATE_MAPS),
 812	fsparam_string	("delegate_progs",		OPT_DELEGATE_PROGS),
 813	fsparam_string	("delegate_attachs",		OPT_DELEGATE_ATTACHS),
 814	{}
 815};
 816
 
 
 
 
 817static int bpf_parse_param(struct fs_context *fc, struct fs_parameter *param)
 818{
 819	struct bpf_mount_opts *opts = fc->s_fs_info;
 820	struct fs_parse_result result;
 821	kuid_t uid;
 822	kgid_t gid;
 823	int opt, err;
 824
 825	opt = fs_parse(fc, bpf_fs_parameters, param, &result);
 826	if (opt < 0) {
 827		/* We might like to report bad mount options here, but
 828		 * traditionally we've ignored all mount options, so we'd
 829		 * better continue to ignore non-existing options for bpf.
 830		 */
 831		if (opt == -ENOPARAM) {
 832			opt = vfs_parse_fs_param_source(fc, param);
 833			if (opt != -ENOPARAM)
 834				return opt;
 835
 836			return 0;
 837		}
 838
 839		if (opt < 0)
 840			return opt;
 841	}
 842
 843	switch (opt) {
 844	case OPT_UID:
 845		uid = make_kuid(current_user_ns(), result.uint_32);
 846		if (!uid_valid(uid))
 847			goto bad_value;
 848
 849		/*
 850		 * The requested uid must be representable in the
 851		 * filesystem's idmapping.
 852		 */
 853		if (!kuid_has_mapping(fc->user_ns, uid))
 854			goto bad_value;
 855
 856		opts->uid = uid;
 857		break;
 858	case OPT_GID:
 859		gid = make_kgid(current_user_ns(), result.uint_32);
 860		if (!gid_valid(gid))
 861			goto bad_value;
 862
 863		/*
 864		 * The requested gid must be representable in the
 865		 * filesystem's idmapping.
 866		 */
 867		if (!kgid_has_mapping(fc->user_ns, gid))
 868			goto bad_value;
 869
 870		opts->gid = gid;
 871		break;
 872	case OPT_MODE:
 873		opts->mode = result.uint_32 & S_IALLUGO;
 874		break;
 875	case OPT_DELEGATE_CMDS:
 876	case OPT_DELEGATE_MAPS:
 877	case OPT_DELEGATE_PROGS:
 878	case OPT_DELEGATE_ATTACHS: {
 879		struct bpffs_btf_enums info;
 880		const struct btf_type *enum_t;
 881		const char *enum_pfx;
 882		u64 *delegate_msk, msk = 0;
 883		char *p;
 884		int val;
 885
 886		/* ignore errors, fallback to hex */
 887		(void)find_bpffs_btf_enums(&info);
 888
 889		switch (opt) {
 890		case OPT_DELEGATE_CMDS:
 891			delegate_msk = &opts->delegate_cmds;
 892			enum_t = info.cmd_t;
 893			enum_pfx = "BPF_";
 894			break;
 895		case OPT_DELEGATE_MAPS:
 896			delegate_msk = &opts->delegate_maps;
 897			enum_t = info.map_t;
 898			enum_pfx = "BPF_MAP_TYPE_";
 899			break;
 900		case OPT_DELEGATE_PROGS:
 901			delegate_msk = &opts->delegate_progs;
 902			enum_t = info.prog_t;
 903			enum_pfx = "BPF_PROG_TYPE_";
 904			break;
 905		case OPT_DELEGATE_ATTACHS:
 906			delegate_msk = &opts->delegate_attachs;
 907			enum_t = info.attach_t;
 908			enum_pfx = "BPF_";
 909			break;
 910		default:
 911			return -EINVAL;
 912		}
 913
 914		while ((p = strsep(&param->string, ":"))) {
 915			if (strcmp(p, "any") == 0) {
 916				msk |= ~0ULL;
 917			} else if (find_btf_enum_const(info.btf, enum_t, enum_pfx, p, &val)) {
 918				msk |= 1ULL << val;
 919			} else {
 920				err = kstrtou64(p, 0, &msk);
 921				if (err)
 922					return err;
 923			}
 924		}
 925
 926		/* Setting delegation mount options requires privileges */
 927		if (msk && !capable(CAP_SYS_ADMIN))
 928			return -EPERM;
 929
 930		*delegate_msk |= msk;
 931		break;
 932	}
 933	default:
 934		/* ignore unknown mount options */
 935		break;
 936	}
 937
 938	return 0;
 939bad_value:
 940	return invalfc(fc, "Bad value for '%s'", param->key);
 941}
 942
 943struct bpf_preload_ops *bpf_preload_ops;
 944EXPORT_SYMBOL_GPL(bpf_preload_ops);
 945
 946static bool bpf_preload_mod_get(void)
 947{
 948	/* If bpf_preload.ko wasn't loaded earlier then load it now.
 949	 * When bpf_preload is built into vmlinux the module's __init
 950	 * function will populate it.
 951	 */
 952	if (!bpf_preload_ops) {
 953		request_module("bpf_preload");
 954		if (!bpf_preload_ops)
 955			return false;
 956	}
 957	/* And grab the reference, so the module doesn't disappear while the
 958	 * kernel is interacting with the kernel module and its UMD.
 959	 */
 960	if (!try_module_get(bpf_preload_ops->owner)) {
 961		pr_err("bpf_preload module get failed.\n");
 962		return false;
 963	}
 964	return true;
 965}
 966
 967static void bpf_preload_mod_put(void)
 968{
 969	if (bpf_preload_ops)
 970		/* now user can "rmmod bpf_preload" if necessary */
 971		module_put(bpf_preload_ops->owner);
 972}
 973
 974static DEFINE_MUTEX(bpf_preload_lock);
 975
 976static int populate_bpffs(struct dentry *parent)
 977{
 978	struct bpf_preload_info objs[BPF_PRELOAD_LINKS] = {};
 979	int err = 0, i;
 980
 981	/* grab the mutex to make sure the kernel interactions with bpf_preload
 982	 * are serialized
 983	 */
 984	mutex_lock(&bpf_preload_lock);
 985
 986	/* if bpf_preload.ko wasn't built into vmlinux then load it */
 987	if (!bpf_preload_mod_get())
 988		goto out;
 989
 990	err = bpf_preload_ops->preload(objs);
 991	if (err)
 992		goto out_put;
 993	for (i = 0; i < BPF_PRELOAD_LINKS; i++) {
 994		bpf_link_inc(objs[i].link);
 995		err = bpf_iter_link_pin_kernel(parent,
 996					       objs[i].link_name, objs[i].link);
 997		if (err) {
 998			bpf_link_put(objs[i].link);
 999			goto out_put;
1000		}
1001	}
1002out_put:
1003	bpf_preload_mod_put();
1004out:
1005	mutex_unlock(&bpf_preload_lock);
1006	return err;
1007}
1008
1009static int bpf_fill_super(struct super_block *sb, struct fs_context *fc)
1010{
1011	static const struct tree_descr bpf_rfiles[] = { { "" } };
1012	struct bpf_mount_opts *opts = sb->s_fs_info;
1013	struct inode *inode;
1014	int ret;
1015
1016	/* Mounting an instance of BPF FS requires privileges */
1017	if (fc->user_ns != &init_user_ns && !capable(CAP_SYS_ADMIN))
1018		return -EPERM;
1019
1020	ret = simple_fill_super(sb, BPF_FS_MAGIC, bpf_rfiles);
1021	if (ret)
1022		return ret;
1023
1024	sb->s_op = &bpf_super_ops;
1025
1026	inode = sb->s_root->d_inode;
1027	inode->i_uid = opts->uid;
1028	inode->i_gid = opts->gid;
1029	inode->i_op = &bpf_dir_iops;
1030	inode->i_mode &= ~S_IALLUGO;
1031	populate_bpffs(sb->s_root);
1032	inode->i_mode |= S_ISVTX | opts->mode;
1033	return 0;
1034}
1035
1036static int bpf_get_tree(struct fs_context *fc)
1037{
1038	return get_tree_nodev(fc, bpf_fill_super);
1039}
1040
1041static void bpf_free_fc(struct fs_context *fc)
1042{
1043	kfree(fc->s_fs_info);
1044}
1045
1046static const struct fs_context_operations bpf_context_ops = {
1047	.free		= bpf_free_fc,
1048	.parse_param	= bpf_parse_param,
1049	.get_tree	= bpf_get_tree,
1050};
1051
1052/*
1053 * Set up the filesystem mount context.
1054 */
1055static int bpf_init_fs_context(struct fs_context *fc)
1056{
1057	struct bpf_mount_opts *opts;
1058
1059	opts = kzalloc(sizeof(struct bpf_mount_opts), GFP_KERNEL);
1060	if (!opts)
1061		return -ENOMEM;
1062
1063	opts->mode = S_IRWXUGO;
1064	opts->uid = current_fsuid();
1065	opts->gid = current_fsgid();
1066
1067	/* start out with no BPF token delegation enabled */
1068	opts->delegate_cmds = 0;
1069	opts->delegate_maps = 0;
1070	opts->delegate_progs = 0;
1071	opts->delegate_attachs = 0;
1072
1073	fc->s_fs_info = opts;
1074	fc->ops = &bpf_context_ops;
1075	return 0;
1076}
1077
1078static void bpf_kill_super(struct super_block *sb)
1079{
1080	struct bpf_mount_opts *opts = sb->s_fs_info;
1081
1082	kill_litter_super(sb);
1083	kfree(opts);
1084}
1085
1086static struct file_system_type bpf_fs_type = {
1087	.owner		= THIS_MODULE,
1088	.name		= "bpf",
1089	.init_fs_context = bpf_init_fs_context,
1090	.parameters	= bpf_fs_parameters,
1091	.kill_sb	= bpf_kill_super,
1092	.fs_flags	= FS_USERNS_MOUNT,
1093};
1094
1095static int __init bpf_init(void)
1096{
1097	int ret;
1098
1099	ret = sysfs_create_mount_point(fs_kobj, "bpf");
1100	if (ret)
1101		return ret;
1102
1103	ret = register_filesystem(&bpf_fs_type);
1104	if (ret)
1105		sysfs_remove_mount_point(fs_kobj, "bpf");
1106
1107	return ret;
1108}
1109fs_initcall(bpf_init);